Head-to-head comparison

Claude Sonnet 5.5 vs. GPT-5.4

Compare published benchmark results from matching versions and cohorts, API pricing, and technical specifications. Any documented test setup differences remain visible.

shared result series: 135
distinct benchmarks: 10
latest retrieval: 2026-10-15

First modelSecond model

Specifications

General specifications

Release date, availability, context window, architecture, and published parameter counts at a glance. Undisclosed details remain marked as unavailable.

Specifications for Claude Sonnet 5.5 and GPT-5.4
Attribute	Claude Sonnet 5.5	GPT-5.4
Provider	Anthropic	OpenAI
Released	Sep 28, 2026	Mar 5, 2026
Availability	Active	Active
Model type	Proprietary	Proprietary
Parameters	Not published	Not published
Architecture	Not published	Not published
Context window	1,000,000 tokens	1,050,000 tokens
Knowledge cutoff	June 2026	August 31, 2025
Technical sources	Model	Model

Pricing

API pricing compared

The table compares official standard API rates per 1M tokens, including published cache pricing. Batch, fast, priority, regional surcharges, tool calls, and cloud platform pricing are excluded.

API pricing for Claude Sonnet 5.5 and GPT-5.4
Price type	Claude Sonnet 5.5	GPT-5.4
API input	$2	$2.5 ≤272K / $5 >272K
API output	$10	$15 ≤272K / $22.5 >272K
Cache read	$0.1	$0.25 ≤272K / $0.5 >272K
Cache write (5 min.)	$2.5	Not listed
Cache write (1 hr.)	$4	Not listed
Price verified	10/13/2026 Source	09/22/2026 Source

Claude Sonnet 5.5GPT-5.4

API inputUSD per 1M tokens, base tier

Claude Sonnet 5.5$2

GPT-5.4$2.5

API outputUSD per 1M tokens, base tier

Claude Sonnet 5.5$10

GPT-5.4$15

Cache readUSD per 1M tokens, base tier

Claude Sonnet 5.5$0.1

GPT-5.4$0.25

Key to the marks:better valuebetter, but not conclusivetie

Pricing by provider

Direct developer pricing and provider endpoint tariffs routed through OpenRouter are listed separately. Prices are in USD per 1M tokens. Context-dependent direct tiers plus regional, Flex, Priority, and other OpenRouter tariffs remain individually identifiable.

Provider pricing for Claude Sonnet 5.5 and GPT-5.4
Model	Provider	Purchase route and tariff	Input per 1M tokens	Output per 1M tokens	Source
Claude Sonnet 5.5	Anthropic	Direct from developerStandard API	Lowest listed price: $2	Lowest listed price: $10	Source10/13/2026
	Amazon Bedrock	Via OpenRouteramazon-bedrock	Lowest listed price: $2	Lowest listed price: $10	Source10/17/2026
	Anthropic	Via OpenRouteranthropic	Lowest listed price: $2	Lowest listed price: $10	Source10/17/2026
	Azure	Via OpenRouterazure/global	Lowest listed price: $2	Lowest listed price: $10	Source10/17/2026
	Azure	Via OpenRouterazure/us	$2.2	$11	Source10/17/2026
	Claude Platform on AWS	Via OpenRouterclaude-on-aws	Lowest listed price: $2	Lowest listed price: $10	Source10/17/2026
	Google	Via OpenRoutergoogle-vertex/europe	$2.2	$11	Source10/17/2026
	Google	Via OpenRoutergoogle-vertex/global	Lowest listed price: $2	Lowest listed price: $10	Source10/17/2026
	Google	Via OpenRoutergoogle-vertex/us	$2.2	$11	Source10/17/2026
GPT-5.4	OpenAI	Direct from developerUp to 272,000 context tokens	$2.5	$15	Source09/22/2026
	OpenAI	Direct from developerAbove 272,000 context tokens	$5	$22.5	Source09/22/2026
	Amazon Bedrock	Via OpenRouteramazon-bedrock/us-east-1	$2.75	$16.5	Source10/17/2026
	Azure	Via OpenRouterazure	$2.5	$15	Source10/17/2026
	Azure	Via OpenRouterazure/eu	$2.75	$16.5	Source10/17/2026
	Azure	Via OpenRouterazure/us	$2.75	$16.5	Source10/17/2026
	OpenAI	Via OpenRouteropenai	$2.5	$15	Source10/17/2026
	OpenAI	Via OpenRouteropenai/fast	$5	$30	Source10/17/2026
	OpenAI	Via OpenRouteropenai/flex	Lowest listed price: $1.25	Lowest listed price: $7.5	Source10/17/2026

Performance

Shared benchmarks

Only published results with the same benchmark version, task, metric, and comparison cohort are paired. Different reasoning levels, agents, harnesses, or output limits appear directly in the table.

Capability profile from matched benchmarks

Each axis averages directly matched benchmark families. A value of 100 means the stronger value within this pair, not a universal quality score.

Not enough directly matched categories for a reliable radar

Only 0 of at least 5 categories have comparable, direct model measurements. The benchmark tables below show the available evidence without filling missing axes.

Claude Sonnet 5.5GPT-5.4

Code MigrationOverall. Higher is better. Test setup not documented as identical

Claude Sonnet 5.569.825 %

GPT-5.434.984 %

Harvey's Legal Agent BenchmarkOverall · Task fully resolved. Higher is better. Test setup not documented as identical

Claude Sonnet 5.52.917 %

GPT-5.40 %

LMArena Code Arena WebDevOverall. Higher is better. Test setup not documented as identical

Claude Sonnet 5.51,712.984

GPT-5.41,401.766

LMArena Text, Style ControlOverall. Higher is better. Test setup not documented as identical

Claude Sonnet 5.51,475.266

GPT-5.41,464.728

LMArena Vision, Style ControlOverall. Higher is better. Test setup not documented as identical

Claude Sonnet 5.51,266.189

GPT-5.41,278.895

MedCodeOverall. Higher is better. Test setup not documented as identical

Claude Sonnet 5.552.92 %

GPT-5.441.292 %

MedScribeOverall. Higher is better. Test setup not documented as identical

Claude Sonnet 5.591.101 %

GPT-5.477.549 %

ProgramBenchOverall. Higher is better. Test setup not documented as identical

Claude Sonnet 5.56.5 %

GPT-5.40 %

SAGEOverall. Higher is better. Test setup not documented as identical

Claude Sonnet 5.551.772 %

GPT-5.443.312 %

Vibe Code Bench v1.1Overall. Higher is better. Test setup not documented as identical

Claude Sonnet 5.592.392 %

GPT-5.467.421 %

Code MigrationCLI. Higher is better. Test setup not documented as identical

Claude Sonnet 5.569.766 %

GPT-5.423.788 %

Harvey's Legal Agent BenchmarkAntitrust Competition · Criteria passed. Higher is better. Test setup not documented as identical

Claude Sonnet 5.585.145 %

GPT-5.476.449 %

Key to the marks:better valuebetter, but not conclusivetie

Benchmark scores for Claude Sonnet 5.5 and GPT-5.4
Benchmark and task	Claude Sonnet 5.5	GPT-5.4	Source
Code MigrationOverall Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	Higher measured value69.825 %	34.984 %	Vals AI Source details Data as of: 10/12/2026
Harvey's Legal Agent BenchmarkOverall · Task fully resolved Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: task resolution rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	Higher measured value2.917 %	0 %	Vals AI Source details Data as of: 10/12/2026
LMArena Code Arena WebDevOverall Test details Test setup not documented as identicalNo direct winnerVersion: webdevMetric: WebDev Arena Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-high, Reasoning: high, Harness: Arena Code Arena WebDev GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Code Arena WebDevClaude Sonnet 5.5: Arena rating from 4,605 published pairwise votes.GPT-5.4: Arena rating from 3,990 published pairwise votes.	Higher measured value1,712.984 95% confidence interval: ±10.362The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,401.766 95% confidence interval: ±10.455The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Text, Style ControlOverall Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Text, style control GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Text, style controlClaude Sonnet 5.5: Arena rating from 6,254 published pairwise votes.GPT-5.4: Arena rating from 67,880 published pairwise votes.	Higher measured value1,475.266 95% confidence interval: ±7.984The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,464.728 95% confidence interval: ±3.69The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Vision, Style ControlOverall Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Vision, style control GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Vision, style controlClaude Sonnet 5.5: Arena rating from 2,068 published pairwise votes.GPT-5.4: Arena rating from 22,832 published pairwise votes.	1,266.189 95% confidence interval: ±13.713The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	Higher measured value1,278.895 95% confidence interval: ±6.536The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/09/2026
MedCodeOverall Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 30,000 tokens	Higher measured value52.92 %	41.292 %	Vals AI Source details Data as of: 10/13/2026
MedScribeOverall Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 30,000 tokens	Higher measured value91.101 %	77.549 %	Vals AI Source details Data as of: 10/13/2026
ProgramBenchOverall Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: fully resolved rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	Higher measured value6.5 %	0 %	Vals AI Source details Data as of: 10/12/2026
SAGEOverall Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 30,000 tokens	Higher measured value51.772 %	43.312 %	Vals AI Source details Data as of: 10/08/2026
Vibe Code Bench v1.1Overall Test details Test setup not documented as identicalNo direct winnerVersion: 1.1Metric: accuracyScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Harness: OpenHands, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Harness: OpenHands, Output limit: 128,000 tokens	Higher measured value92.392 %	67.421 %	Vals AI Source details Data as of: 10/12/2026
Code MigrationCLI Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	Higher measured value69.766 %	23.788 %	Vals AI Source details Data as of: 10/12/2026
Harvey's Legal Agent BenchmarkAntitrust Competition · Criteria passed Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: criteria pass rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	Higher measured value85.145 %	76.449 %	Vals AI Source details Data as of: 10/12/2026
LMArena Code Arena WebDevOverall Test details Test setup not documented as identicalNo direct winnerVersion: webdevMetric: WebDev Arena Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Code Arena WebDev GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Code Arena WebDevClaude Sonnet 5.5: Arena rating from 2,858 published pairwise votes.GPT-5.4: Arena rating from 3,990 published pairwise votes.	Higher measured value1,773.661 95% confidence interval: ±13.85The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,401.766 95% confidence interval: ±10.455The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Text, Style ControlOverall Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Text, style control GPT-5.4: Snapshot: gpt-5.4-high, Reasoning: high, Harness: Arena Text, style controlClaude Sonnet 5.5: Arena rating from 6,254 published pairwise votes.GPT-5.4: Arena rating from 64,787 published pairwise votes.	Higher measured value1,475.266 95% confidence interval: ±7.984The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,475.066 95% confidence interval: ±3.765The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Vision, Style ControlOverall Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Vision, style control GPT-5.4: Snapshot: gpt-5.4-high, Reasoning: high, Harness: Arena Vision, style controlClaude Sonnet 5.5: Arena rating from 2,068 published pairwise votes.GPT-5.4: Arena rating from 29,881 published pairwise votes.	1,266.189 95% confidence interval: ±13.713The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	Higher measured value1,284.987 95% confidence interval: ±5.834The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/09/2026
ProgramBenchAlmost Resolved Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: almost-resolved rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	Higher measured value49.5 %	7 %	Vals AI Source details Data as of: 10/12/2026
SAGEAutomata Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 30,000 tokens	Higher measured value43.254 %	21.428 %	Vals AI Source details Data as of: 10/08/2026
Code MigrationCOBOL Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	Higher measured value70 %	68.571 %	Vals AI Source details Data as of: 10/12/2026
Harvey's Legal Agent BenchmarkAntitrust Competition · Task fully resolved Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: task resolution rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	No direct winner0 %	No direct winner0 %	Vals AI Source details Data as of: 10/12/2026
LMArena Code Arena WebDevImage to webdev Test details Test setup not documented as identicalNo direct winnerVersion: webdevMetric: WebDev Arena Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Code Arena WebDev GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Code Arena WebDevClaude Sonnet 5.5: Arena rating from 1,074 published pairwise votes.GPT-5.4: Arena rating from 3,076 published pairwise votes.	Higher measured value1,740.047 95% confidence interval: ±19.9The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,440.409 95% confidence interval: ±11.975The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/09/2026
LMArena Text, Style ControlBusiness, management and finance Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Text, style control GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Text, style controlClaude Sonnet 5.5: Arena rating from 1,089 published pairwise votes.GPT-5.4: Arena rating from 13,848 published pairwise votes.	1,466.655 95% confidence interval: ±17.914The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	Higher measured value1,473.882 95% confidence interval: ±6.361The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Vision, Style ControlDiagram Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Vision, style control GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Vision, style controlClaude Sonnet 5.5: Arena rating from 514 published pairwise votes.GPT-5.4: Arena rating from 6,232 published pairwise votes.	Higher measured value1,320.26 95% confidence interval: ±26.586The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,314.348 95% confidence interval: ±9.274The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/09/2026
ProgramBenchFully Resolved Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: fully resolved rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	Higher measured value6.5 %	0 %	Vals AI Source details Data as of: 10/12/2026
SAGECalculus Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 30,000 tokens	Higher measured value78.395 %	70.988 %	Vals AI Source details Data as of: 10/08/2026
Code MigrationCode Quality Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: mean reviewer code-quality ratingScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	Higher measured value4.601 / 5	3.632 / 5	Vals AI Source details Data as of: 10/12/2026
Harvey's Legal Agent BenchmarkArbitration International Dispute Resolution · Criteria passed Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: criteria pass rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	36.918 %	Higher measured value66.667 %	Vals AI Source details Data as of: 10/12/2026
LMArena Code Arena WebDevWebdev Test details Test setup not documented as identicalNo direct winnerVersion: webdevMetric: WebDev Arena Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-high, Reasoning: high, Harness: Arena Code Arena WebDev GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Code Arena WebDevClaude Sonnet 5.5: Arena rating from 4,605 published pairwise votes.GPT-5.4: Arena rating from 3,990 published pairwise votes.	Higher measured value1,712.984 95% confidence interval: ±10.362The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,401.766 95% confidence interval: ±10.455The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Text, Style ControlBusiness, management and finance Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Text, style control GPT-5.4: Snapshot: gpt-5.4-high, Reasoning: high, Harness: Arena Text, style controlClaude Sonnet 5.5: Arena rating from 1,089 published pairwise votes.GPT-5.4: Arena rating from 13,095 published pairwise votes.	1,466.655 95% confidence interval: ±17.914The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	Higher measured value1,481.483 95% confidence interval: ±6.462The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Vision, Style ControlDiagram Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Vision, style control GPT-5.4: Snapshot: gpt-5.4-high, Reasoning: high, Harness: Arena Vision, style controlClaude Sonnet 5.5: Arena rating from 514 published pairwise votes.GPT-5.4: Arena rating from 8,005 published pairwise votes.	1,320.26 95% confidence interval: ±26.586The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	Higher measured value1,321.303 95% confidence interval: ±8.334The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/09/2026
ProgramBenchRaw Pass Rate Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: raw test pass rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	Higher measured value76.304 %	68.596 %	Vals AI Source details Data as of: 10/12/2026
SAGELinear Algebra Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 30,000 tokens	Higher measured value70.792 %	68.826 %	Vals AI Source details Data as of: 10/08/2026
Harvey's Legal Agent BenchmarkArbitration International Dispute Resolution · Task fully resolved Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: task resolution rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	No direct winner0 %	No direct winner0 %	Vals AI Source details Data as of: 10/12/2026
LMArena Code Arena WebDevWebdev Test details Test setup not documented as identicalNo direct winnerVersion: webdevMetric: WebDev Arena Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Code Arena WebDev GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Code Arena WebDevClaude Sonnet 5.5: Arena rating from 2,858 published pairwise votes.GPT-5.4: Arena rating from 3,990 published pairwise votes.	Higher measured value1,773.661 95% confidence interval: ±13.85The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,401.766 95% confidence interval: ±10.455The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Text, Style ControlChinese Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Text, style control GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Text, style controlClaude Sonnet 5.5: Arena rating from 508 published pairwise votes.GPT-5.4: Arena rating from 4,351 published pairwise votes.	Higher measured value1,546.494 95% confidence interval: ±26.454The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,493.612 95% confidence interval: ±9.825The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Vision, Style ControlEnglish Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Vision, style control GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Vision, style controlClaude Sonnet 5.5: Arena rating from 657 published pairwise votes.GPT-5.4: Arena rating from 9,601 published pairwise votes.	1,266.361 95% confidence interval: ±23.51The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	Higher measured value1,277.195 95% confidence interval: ±8.87The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/09/2026
SAGEPartial Differential Equations Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 30,000 tokens	Higher measured value41.27 %	39.683 %	Vals AI Source details Data as of: 10/08/2026
Harvey's Legal Agent BenchmarkBanking Finance · Criteria passed Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: criteria pass rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	Higher measured value89.591 %	80.297 %	Vals AI Source details Data as of: 10/12/2026
LMArena Code Arena WebDevWebdev-html Test details Test setup not documented as identicalNo direct winnerVersion: webdevMetric: WebDev Arena Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-high, Reasoning: high, Harness: Arena Code Arena WebDev GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Code Arena WebDevClaude Sonnet 5.5: Arena rating from 707 published pairwise votes.GPT-5.4: Arena rating from 528 published pairwise votes.	Higher measured value1,710.04 95% confidence interval: ±25.597The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,434.816 95% confidence interval: ±27.294The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Text, Style ControlChinese Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Text, style control GPT-5.4: Snapshot: gpt-5.4-high, Reasoning: high, Harness: Arena Text, style controlClaude Sonnet 5.5: Arena rating from 508 published pairwise votes.GPT-5.4: Arena rating from 4,130 published pairwise votes.	Higher measured value1,546.494 95% confidence interval: ±26.454The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,505.784 95% confidence interval: ±9.998The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Vision, Style ControlEnglish Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Vision, style control GPT-5.4: Snapshot: gpt-5.4-high, Reasoning: high, Harness: Arena Vision, style controlClaude Sonnet 5.5: Arena rating from 657 published pairwise votes.GPT-5.4: Arena rating from 11,832 published pairwise votes.	1,266.361 95% confidence interval: ±23.51The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	Higher measured value1,282.055 95% confidence interval: ±8.176The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/09/2026
Harvey's Legal Agent BenchmarkBanking Finance · Task fully resolved Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: task resolution rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	No direct winner0 %	No direct winner0 %	Vals AI Source details Data as of: 10/12/2026
LMArena Code Arena WebDevWebdev-html Test details Test setup not documented as identicalNo direct winnerVersion: webdevMetric: WebDev Arena Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Code Arena WebDev GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Code Arena WebDevClaude Sonnet 5.5: Arena rating from 423 published pairwise votes.GPT-5.4: Arena rating from 528 published pairwise votes.	Higher measured value1,737.658 95% confidence interval: ±36.636The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,434.816 95% confidence interval: ±27.294The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Text, Style ControlCoding Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Text, style control GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Text, style controlClaude Sonnet 5.5: Arena rating from 1,506 published pairwise votes.GPT-5.4: Arena rating from 18,933 published pairwise votes.	Higher measured value1,526.562 95% confidence interval: ±15.395The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,511.592 95% confidence interval: ±5.788The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Vision, Style ControlHomework Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Vision, style control GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Vision, style controlClaude Sonnet 5.5: Arena rating from 295 published pairwise votes.GPT-5.4: Arena rating from 3,099 published pairwise votes.	Higher measured value1,318.835 95% confidence interval: ±34.105The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,303.012 95% confidence interval: ±11.693The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/09/2026
Harvey's Legal Agent BenchmarkBankruptcy Restructuring · Criteria passed Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: criteria pass rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	59.17 %	Higher measured value68.512 %	Vals AI Source details Data as of: 10/12/2026
LMArena Code Arena WebDevWebdev-react Test details Test setup not documented as identicalNo direct winnerVersion: webdevMetric: WebDev Arena Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-high, Reasoning: high, Harness: Arena Code Arena WebDev GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Code Arena WebDevClaude Sonnet 5.5: Arena rating from 2,976 published pairwise votes.GPT-5.4: Arena rating from 2,727 published pairwise votes.	Higher measured value1,747.964 95% confidence interval: ±12.95The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,391.351 95% confidence interval: ±12.752The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Text, Style ControlCoding Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Text, style control GPT-5.4: Snapshot: gpt-5.4-high, Reasoning: high, Harness: Arena Text, style controlClaude Sonnet 5.5: Arena rating from 1,506 published pairwise votes.GPT-5.4: Arena rating from 17,605 published pairwise votes.	Higher measured value1,526.562 95% confidence interval: ±15.395The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,520.364 95% confidence interval: ±5.837The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Vision, Style ControlHomework Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Vision, style control GPT-5.4: Snapshot: gpt-5.4-high, Reasoning: high, Harness: Arena Vision, style controlClaude Sonnet 5.5: Arena rating from 295 published pairwise votes.GPT-5.4: Arena rating from 3,876 published pairwise votes.	1,318.835 95% confidence interval: ±34.105The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	Higher measured value1,331.465 95% confidence interval: ±10.598The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/09/2026
Harvey's Legal Agent BenchmarkBankruptcy Restructuring · Task fully resolved Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: task resolution rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	No direct winner0 %	No direct winner0 %	Vals AI Source details Data as of: 10/12/2026
LMArena Code Arena WebDevWebdev-react Test details Test setup not documented as identicalNo direct winnerVersion: webdevMetric: WebDev Arena Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Code Arena WebDev GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Code Arena WebDevClaude Sonnet 5.5: Arena rating from 1,812 published pairwise votes.GPT-5.4: Arena rating from 2,727 published pairwise votes.	Higher measured value1,823.885 95% confidence interval: ±17.562The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,391.351 95% confidence interval: ±12.752The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Text, Style ControlCreative writing Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Text, style control GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Text, style controlClaude Sonnet 5.5: Arena rating from 1,405 published pairwise votes.GPT-5.4: Arena rating from 11,357 published pairwise votes.	Higher measured value1,465.364 95% confidence interval: ±16.888The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,434.365 95% confidence interval: ±6.8The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Vision, Style ControlOcr Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Vision, style control GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Vision, style controlClaude Sonnet 5.5: Arena rating from 1,451 published pairwise votes.GPT-5.4: Arena rating from 16,464 published pairwise votes.	1,280.319 95% confidence interval: ±15.866The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	Higher measured value1,290.283 95% confidence interval: ±6.649The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/09/2026
Harvey's Legal Agent BenchmarkCapital Markets · Criteria passed Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: criteria pass rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	Higher measured value90.98 %	81.569 %	Vals AI Source details Data as of: 10/12/2026
LMArena Text, Style ControlCreative writing Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Text, style control GPT-5.4: Snapshot: gpt-5.4-high, Reasoning: high, Harness: Arena Text, style controlClaude Sonnet 5.5: Arena rating from 1,405 published pairwise votes.GPT-5.4: Arena rating from 11,142 published pairwise votes.	Higher measured value1,465.364 95% confidence interval: ±16.888The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,442.006 95% confidence interval: ±6.936The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
LMArena Vision, Style ControlOcr Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Vision, style control GPT-5.4: Snapshot: gpt-5.4-high, Reasoning: high, Harness: Arena Vision, style controlClaude Sonnet 5.5: Arena rating from 1,451 published pairwise votes.GPT-5.4: Arena rating from 21,216 published pairwise votes.	1,280.319 95% confidence interval: ±15.866The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	Higher measured value1,299.535 95% confidence interval: ±5.805The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/09/2026
Harvey's Legal Agent BenchmarkCapital Markets · Task fully resolved Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: task resolution rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	No direct winner0 %	No direct winner0 %	Vals AI Source details Data as of: 10/12/2026
LMArena Text, Style ControlEnglish Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Text, style control GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Text, style controlClaude Sonnet 5.5: Arena rating from 2,393 published pairwise votes.GPT-5.4: Arena rating from 30,866 published pairwise votes.	Higher measured value1,490.413 95% confidence interval: ±12.395The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,468.033 95% confidence interval: ±4.897The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
Harvey's Legal Agent BenchmarkCorporate Governance · Criteria passed Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: criteria pass rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	50.685 %	Higher measured value72.945 %	Vals AI Source details Data as of: 10/12/2026
LMArena Text, Style ControlEnglish Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Text, style control GPT-5.4: Snapshot: gpt-5.4-high, Reasoning: high, Harness: Arena Text, style controlClaude Sonnet 5.5: Arena rating from 2,393 published pairwise votes.GPT-5.4: Arena rating from 29,345 published pairwise votes.	Higher measured value1,490.413 95% confidence interval: ±12.395The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,476.182 95% confidence interval: ±4.944The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
Harvey's Legal Agent BenchmarkCorporate Governance · Task fully resolved Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: task resolution rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	No direct winner0 %	No direct winner0 %	Vals AI Source details Data as of: 10/12/2026
LMArena Text, Style ControlEntertainment, sports and media Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Text, style control GPT-5.4: Snapshot: gpt-5.4, Harness: Arena Text, style controlClaude Sonnet 5.5: Arena rating from 1,576 published pairwise votes.GPT-5.4: Arena rating from 14,807 published pairwise votes.	Higher measured value1,460.568 95% confidence interval: ±15.574The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,431.662 95% confidence interval: ±6.25The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
Harvey's Legal Agent BenchmarkCorporate Ma · Criteria passed Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: criteria pass rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	75.962 %	Higher measured value81.731 %	Vals AI Source details Data as of: 10/12/2026
LMArena Text, Style ControlEntertainment, sports and media Test details Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Sonnet 5.5: Snapshot: claude-sonnet-5.5-xhigh, Reasoning: xhigh, Harness: Arena Text, style control GPT-5.4: Snapshot: gpt-5.4-high, Reasoning: high, Harness: Arena Text, style controlClaude Sonnet 5.5: Arena rating from 1,576 published pairwise votes.GPT-5.4: Arena rating from 14,168 published pairwise votes.	Higher measured value1,460.568 95% confidence interval: ±15.574The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	1,440.873 95% confidence interval: ±6.398The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.	LMArena Source details Data as of: 10/13/2026
Harvey's Legal Agent BenchmarkCorporate Ma · Task fully resolved Test details Test setup not documented as identicalNo direct winnerVersion: 1Metric: task resolution rateScoring: Higher is betterClaude Sonnet 5.5: Compute: max, Temperature: 1, Output limit: 128,000 tokens GPT-5.4: Snapshot: gpt-5.4-2026-03-05, Reasoning: xhigh, Output limit: 128,000 tokens	No direct winner0 %	No direct winner0 %	Vals AI Source details Data as of: 10/12/2026
Key to the marks:better valuebetter, but not conclusivetie

Our model tests

Our tests

SVG short test

One SVG prompt and fixed output settings. Captures are shown only with 0 reported reasoning tokens or documented lack of reasoning controls. Identical test conditions do not make the models' internal compute identical.

Shared test round: 4ae5d8eeb9a9

Claude Sonnet 5.5

Unavailable in this round

GPT-5.4

Unavailable in this round

Earlier tests

Pelican on a bicycle

A community classic for free-form SVG drawing.

Claude Sonnet 5.5

Reasoning (requested): high

Current attempt

Run date: Oct 1, 2026
API provider: openrouter
Requested API model: anthropic/claude-sonnet-5.5
API model reported in response: anthropic/claude-sonnet-5.5
Reasoning setting: high
Pinned endpoint: Anthropic | anthropic/claude-sonnet-5.5-20260928
Temperature: Provider default
Token limit including reasoning: 8,192 tokens
Test protocol: community-visual-tests-v1
Technical checks: Passed, not a quality rating
Response time: 28.2 s

Visually reviewed. This individual test does not establish overall model quality.

GPT-5.4

Reasoning (requested): high

Current attempt

Run date: Oct 1, 2026
API provider: openrouter
Requested API model: openai/gpt-5.4
API model reported in response: openai/gpt-5.4
Reasoning setting: high
Pinned endpoint: OpenAI | openai/gpt-5.4-20260305
Temperature: Provider default
Token limit including reasoning: 8,192 tokens
Test protocol: community-visual-tests-v1
Technical checks: Passed, not a quality rating
Response time: 116.8 s

Visually reviewed. This individual test does not establish overall model quality.

Earlier attempts (1)

Sep 7, 2026API request failedReasoning setting: high

Prompt and test conditions

Generate an SVG of a pelican riding a bicycle

Token limit including reasoning: 8,192 tokens

Task origin (Simon Willison)

How to read this comparison

Winning one benchmark is not an overall verdict

Your workload, budget, and required context length matter most. A coding benchmark says little about visual understanding or agent performance.

The standard price table uses direct developer rates. The provider comparison labels OpenRouter endpoints, regions, and special tariffs separately.

Every result links to its measurement source and retrieval date. Labels distinguish vendor reports, official benchmark leaderboards, and independent evaluations. Disputed or archived results are excluded.

More matchups

Compare Claude Sonnet 5.5 and GPT-5.4 with other leading models using the same data and benchmark logic.

View the complete LLM comparison

Claude Sonnet 5.5 vs. GPT-5.4

Change model selection

General specifications

API pricing compared

Shared benchmarks

Capability profile from matched benchmarks

Our tests

SVG short test

Claude Sonnet 5.5

GPT-5.4

Pelican on a bicycle

Winning one benchmark is not an overall verdict

Related LLM comparisons