Skip to main content

Head-to-head comparison

Claude Opus 4.6 vs. GPT-5.6 Terra

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

shared result series
24
distinct benchmark metrics
9
latest retrieval
2026-10-19

Change model selection

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 Opus 4.6 and GPT-5.6 Terra
AttributeClaude Opus 4.6GPT-5.6 Terra
ProviderAnthropicOpenAI
ReleasedFeb 2026Jul 9, 2026
AvailabilityActiveActive
Model typeProprietaryProprietary
ParametersNot publishedNot published
ArchitectureNot publishedNot published
Context window1,000,000 tokens1,050,000 tokens
Knowledge cutoffMay 2025February 16, 2026
Technical sourcesModel, ContextModel, Context, Knowledge cutoff

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 Opus 4.6 and GPT-5.6 Terra
Price typeClaude Opus 4.6GPT-5.6 Terra
API input$5$2 ≤272K / $4 >272K
API output$25$12 ≤272K / $18 >272K
Cache read$0.5$0.2 ≤272K / $0.4 >272K
Cache write (5 min.)$6.25$2.5 ≤272K / $5 >272K
Cache write (1 hr.)$10Not listed
Price verified09/22/2026 Source09/22/2026 Source
Claude Opus 4.6GPT-5.6 Terra
API inputUSD per 1M tokens, base tier
Claude Opus 4.6$5
GPT-5.6 TerraLower price$2
API outputUSD per 1M tokens, base tier
Claude Opus 4.6$25
GPT-5.6 TerraLower price$12
Cache readUSD per 1M tokens, base tier
Claude Opus 4.6$0.5
GPT-5.6 TerraLower price$0.2

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 Opus 4.6 and GPT-5.6 Terra
ModelProviderPurchase route and tariffInput per 1M tokensOutput per 1M tokensSource
Claude Opus 4.6AnthropicDirect from developerStandard APILowest listed price: $5Lowest listed price: $25Source09/22/2026
Amazon BedrockVia OpenRouteramazon-bedrockLowest listed price: $5Lowest listed price: $25Source10/26/2026
AnthropicVia OpenRouteranthropicLowest listed price: $5Lowest listed price: $25Source10/26/2026
AzureVia OpenRouterazure/globalLowest listed price: $5Lowest listed price: $25Source10/26/2026
Claude Platform on AWSVia OpenRouterclaude-on-awsLowest listed price: $5Lowest listed price: $25Source10/26/2026
GoogleVia OpenRoutergoogle-vertex/europe$5.5$27.5Source10/26/2026
GoogleVia OpenRoutergoogle-vertex/globalLowest listed price: $5Lowest listed price: $25Source10/26/2026
GPT-5.6 TerraOpenAIDirect from developerUp to 272,000 context tokens$2$12Source09/22/2026
OpenAIDirect from developerAbove 272,000 context tokens$4$18Source09/22/2026
Amazon BedrockVia OpenRouteramazon-bedrock/us-east-1$2.2$13.2Source10/26/2026
AzureVia OpenRouterazure$2$12Source10/26/2026
AzureVia OpenRouterazure/eu$2.2$13.2Source10/26/2026
AzureVia OpenRouterazure/us$2.2$13.2Source10/26/2026
OpenAIVia OpenRouteropenai$2$12Source10/26/2026
OpenAIVia OpenRouteropenai/fast$4$24Source10/26/2026
OpenAIVia OpenRouteropenai/flexLowest listed price: $1Lowest listed price: $6Source10/26/2026

Benchmark index

Benchmark index compared

The index combines independent benchmarks on a scale from 0 to 100. A lead only counts when a model comes first in at least 95% of 200 recalculations with randomly redrawn test families. Otherwise both count as level.

Open the AI leaderboard
Claude Opus 4.6

Overall index

No overall score yet

GPT-5.6 Terra

Overall index

52.1

Claude Opus 4.6 leads in 3 of 5 shared areas.

AreaClaude Opus 4.6GPT-5.6 Terra
Coding43.861.3
Agents48.061.8
Finance72.960.4
Legal71.541.4
Health63.040.3

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

Claude Opus 4.6GPT-5.6 Terra
AA-Omniscience AccuracyOverall. Higher is better. No documented setup difference
Claude Opus 4.6Winner47.0%
GPT-5.6 Terra46.8%
AA-Omniscience Hallucination RateOverall. Lower is better. No documented setup difference
Claude Opus 4.6Winner62.8%
GPT-5.6 Terra87.9%
APEX-Agents-AAOverall. Higher is better. No documented setup difference
Claude Opus 4.633.0%
GPT-5.6 TerraWinner38.9%
IFBenchOverall. Higher is better. No documented setup difference
Claude Opus 4.653.1%
GPT-5.6 TerraWinner71.2%
LMArena Vision, Style ControlOverall. Higher is better. Test setup not documented as identical
Claude Opus 4.6Higher measured value1,294.46
GPT-5.6 Terra1,268.044
MedCodeOverall. Higher is better. Test setup not documented as identical
Claude Opus 4.6Higher measured value48.244 %
GPT-5.6 Terra43.414 %
MedScribeOverall. Higher is better. Test setup not documented as identical
Claude Opus 4.6Higher measured value86.738 %
GPT-5.6 Terra82.869 %
MMMU-ProOverall. Higher is better. No documented setup difference
Claude Opus 4.675.4%
GPT-5.6 TerraWinner80.7%
Vibe Code Bench v1.1Overall. Higher is better. Test setup not documented as identical
Claude Opus 4.657.573 %
GPT-5.6 TerraHigher measured value74.594 %
LMArena Vision, Style ControlOverall. Higher is better. Test setup not documented as identical
Claude Opus 4.6Higher measured value1,299.232
GPT-5.6 Terra1,268.044
LMArena Vision, Style ControlChinese. Higher is better. Test setup not documented as identical
Claude Opus 4.6Higher measured value1,348.115
GPT-5.6 Terra1,313.871
LMArena Vision, Style ControlChinese. Higher is better. Test setup not documented as identical
Claude Opus 4.6Higher measured value1,351.953
GPT-5.6 Terra1,313.871

Key to the marks:better valuebetter, but not conclusivetie

Benchmark scores for Claude Opus 4.6 and GPT-5.6 Terra
Benchmark and taskClaude Opus 4.6GPT-5.6 TerraSource
AA-Omniscience AccuracyOverall
Test details
No documented difference in the test setupVersion: AA Omniscience diagnosticMetric: AA-Omniscience AccuracyScoring: Higher is betterClaude Opus 4.6: Snapshot: Claude Opus 4.6 (Max), Reasoning: Max, Attempts: AA-reported aggregate; diagnostic-specific run count is not stated on the leaderboard
GPT-5.6 Terra: Snapshot: GPT-5.6 Terra (Max), Reasoning: Max, Attempts: AA-reported aggregate; diagnostic-specific run count is not stated on the leaderboardClaude Opus 4.6: Exact AA configuration: Claude Opus 4.6 (Max) (release claude-opus-4-6, row claude-opus-4-6-adaptive, UUID 53c98840-47af-49aa-94e6-469fb17e9a1b, reasoning=true, source reasoning-token configuration budget=0, suffix=Max). This configured value is not measured token consumption. Proportion of all questions answered correctly, regardless of whether the model answers or abstains on other questions.GPT-5.6 Terra: Exact AA configuration: GPT-5.6 Terra (Max) (release gpt-5-6-terra, row gpt-5-6-terra, UUID bcf8db0a-3bb6-4d82-9516-0f57370c85a6, reasoning=true, source reasoning-token configuration budget=0, suffix=Max). This configured value is not measured token consumption. Proportion of all questions answered correctly, regardless of whether the model answers or abstains on other questions.
Winner47.0%
This is the separately published accuracy diagnostic, not the AA-Omniscience Index. AA identifies higher values as better.
46.8%
This is the separately published accuracy diagnostic, not the AA-Omniscience Index. AA identifies higher values as better.		Artificial Analysis
Source details
Data as of: 10/19/2026
AA-Omniscience Hallucination RateOverall
Test details
No documented difference in the test setupVersion: AA Omniscience diagnosticMetric: AA-Omniscience Hallucination RateScoring: Lower is betterClaude Opus 4.6: Snapshot: Claude Opus 4.6 (Max), Reasoning: Max, Attempts: AA-reported aggregate; diagnostic-specific run count is not stated on the leaderboard
GPT-5.6 Terra: Snapshot: GPT-5.6 Terra (Max), Reasoning: Max, Attempts: AA-reported aggregate; diagnostic-specific run count is not stated on the leaderboardClaude Opus 4.6: Exact AA configuration: Claude Opus 4.6 (Max) (release claude-opus-4-6, row claude-opus-4-6-adaptive, UUID 53c98840-47af-49aa-94e6-469fb17e9a1b, reasoning=true, source reasoning-token configuration budget=0, suffix=Max). This configured value is not measured token consumption. Proportion of incorrect answers among all non-correct responses: incorrect / (incorrect + partial answers + not attempted).GPT-5.6 Terra: Exact AA configuration: GPT-5.6 Terra (Max) (release gpt-5-6-terra, row gpt-5-6-terra, UUID bcf8db0a-3bb6-4d82-9516-0f57370c85a6, reasoning=true, source reasoning-token configuration budget=0, suffix=Max). This configured value is not measured token consumption. Proportion of incorrect answers among all non-correct responses: incorrect / (incorrect + partial answers + not attempted).
Winner62.8%
This is the separately published hallucination-rate diagnostic, not the AA-Omniscience Index. AA identifies lower values as better.
87.9%
This is the separately published hallucination-rate diagnostic, not the AA-Omniscience Index. AA identifies lower values as better.		Artificial Analysis
Source details
Data as of: 10/19/2026
APEX-Agents-AAOverall
Test details
No documented difference in the test setupVersion: AA public cohort: 452 tasksMetric: Pass@1 success rateScoring: Higher is betterClaude Opus 4.6: Snapshot: Claude Opus 4.6 (Max), Reasoning: Max, Agent: Stirrup agent, Harness: Artificial Analysis Stirrup, Attempts: pass@1, averaged over 3 repeats per task
GPT-5.6 Terra: Snapshot: GPT-5.6 Terra (Max), Reasoning: Max, Agent: Stirrup agent, Harness: Artificial Analysis Stirrup, Attempts: pass@1, averaged over 3 repeats per taskClaude Opus 4.6: Exact AA configuration: Claude Opus 4.6 (Max) (release claude-opus-4-6, row claude-opus-4-6-adaptive, UUID 53c98840-47af-49aa-94e6-469fb17e9a1b, reasoning=true, source reasoning-token configuration budget=0, suffix=Max). This configured value is not measured token consumption. 452 tasks from the public 480-task release, with three repeats per task. A run passes only if every rubric criterion is satisfied; the leaderboard averages pass@1 across repeats.GPT-5.6 Terra: Exact AA configuration: GPT-5.6 Terra (Max) (release gpt-5-6-terra, row gpt-5-6-terra, UUID bcf8db0a-3bb6-4d82-9516-0f57370c85a6, reasoning=true, source reasoning-token configuration budget=0, suffix=Max). This configured value is not measured token consumption. 452 tasks from the public 480-task release, with three repeats per task. A run passes only if every rubric criterion is satisfied; the leaderboard averages pass@1 across repeats.
33.0%
Two Investment Banking worlds are excluded because they depend on external runtime APIs.
Winner38.9%
Two Investment Banking worlds are excluded because they depend on external runtime APIs.		Artificial Analysis
Source details
Data as of: 10/19/2026
IFBenchOverall
Test details
No documented difference in the test setupVersion: AA test set: 294 questionsMetric: IFBench prompt-level loose accuracy (pass@1)Scoring: Higher is betterClaude Opus 4.6: Snapshot: Claude Opus 4.6 (Max), Reasoning: Max, Attempts: pass@1, 5 repeats per question
GPT-5.6 Terra: Snapshot: GPT-5.6 Terra (Max), Reasoning: Max, Attempts: pass@1, 5 repeats per questionClaude Opus 4.6: Exact AA configuration: Claude Opus 4.6 (Max) (release claude-opus-4-6, row claude-opus-4-6-adaptive, UUID 53c98840-47af-49aa-94e6-469fb17e9a1b, reasoning=true, source reasoning-token configuration budget=0, suffix=Max). This configured value is not measured token consumption. Artificial Analysis runs 294 questions from the IFBench test set with five repeats per question. The official evaluator scores prompt-level accuracy in loose mode as pass@1.GPT-5.6 Terra: Exact AA configuration: GPT-5.6 Terra (Max) (release gpt-5-6-terra, row gpt-5-6-terra, UUID bcf8db0a-3bb6-4d82-9516-0f57370c85a6, reasoning=true, source reasoning-token configuration budget=0, suffix=Max). This configured value is not measured token consumption. Artificial Analysis runs 294 questions from the IFBench test set with five repeats per question. The official evaluator scores prompt-level accuracy in loose mode as pass@1.
53.1%
The often cited 58 refers to verifiable out-of-domain constraint types, not the number of questions in AA's test run.
Winner71.2%
The often cited 58 refers to verifiable out-of-domain constraint types, not the number of questions in AA's test run.		Artificial Analysis
Source details
Data as of: 10/19/2026
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 Opus 4.6: Snapshot: claude-opus-4-6, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 26,853 published pairwise votes.GPT-5.6 Terra: Arena rating from 11,020 published pairwise votes.
Higher measured value1,294.46
95% confidence interval: ±6.329The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,268.044
95% confidence interval: ±7.452The 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 Opus 4.6: Compute: max, Temperature: 1, Output limit: 30,000 tokens
GPT-5.6 Terra: Reasoning: xhigh, Output limit: 30,000 tokens
Higher measured value48.244 %
43.414 %
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 Opus 4.6: Compute: max, Temperature: 1, Output limit: 30,000 tokens
GPT-5.6 Terra: Reasoning: xhigh, Output limit: 30,000 tokens
Higher measured value86.738 %
82.869 %
Vals AI
Source details
Data as of: 10/13/2026
MMMU-ProOverall
Test details
No documented difference in the test setupVersion: AA cohort: 1,730 questionsMetric: pass@1 accuracyScoring: Higher is betterClaude Opus 4.6: Snapshot: Claude Opus 4.6 (Max), Reasoning: Max, Attempts: pass@1, one run per question
GPT-5.6 Terra: Snapshot: GPT-5.6 Terra (Max), Reasoning: Max, Attempts: pass@1, one run per questionClaude Opus 4.6: Exact AA configuration: Claude Opus 4.6 (Max) (release claude-opus-4-6, row claude-opus-4-6-adaptive, UUID 53c98840-47af-49aa-94e6-469fb17e9a1b, reasoning=true, source reasoning-token configuration budget=0, suffix=Max). This configured value is not measured token consumption. The AA run covers 1,730 questions with ten answer options. Answers are extracted by regex and scored pass@1.GPT-5.6 Terra: Exact AA configuration: GPT-5.6 Terra (Max) (release gpt-5-6-terra, row gpt-5-6-terra, UUID bcf8db0a-3bb6-4d82-9516-0f57370c85a6, reasoning=true, source reasoning-token configuration budget=0, suffix=Max). This configured value is not measured token consumption. The AA run covers 1,730 questions with ten answer options. Answers are extracted by regex and scored pass@1.
75.4%
The benchmark page describes 3,460 questions in the full MMMU-Pro dataset; AA's methodology specifies a 1,730-question run.
Winner80.7%
The benchmark page describes 3,460 questions in the full MMMU-Pro dataset; AA's methodology specifies a 1,730-question run.		Artificial Analysis
Source details
Data as of: 10/19/2026
Vibe Code Bench v1.1Overall
Test details
Test setup not documented as identicalNo direct winnerVersion: 1.1Metric: accuracyScoring: Higher is betterClaude Opus 4.6: Compute: max, Harness: OpenHands, Temperature: 1, Output limit: 128,000 tokens
GPT-5.6 Terra: Reasoning: max, Harness: OpenHands, Output limit: 128,000 tokens
57.573 %
Higher measured value74.594 %
Vals AI
Source details
Data as of: 10/12/2026
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 Opus 4.6: Snapshot: claude-opus-4-6-high, Reasoning: high, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 22,304 published pairwise votes.GPT-5.6 Terra: Arena rating from 11,020 published pairwise votes.
Higher measured value1,299.232
95% confidence interval: ±6.456The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,268.044
95% confidence interval: ±7.452The 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 Vision, Style ControlChinese
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Opus 4.6: Snapshot: claude-opus-4-6, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 1,685 published pairwise votes.GPT-5.6 Terra: Arena rating from 642 published pairwise votes.
Higher measured value1,348.115
95% confidence interval: ±17.347The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,313.871
95% confidence interval: ±24.287The 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 Vision, Style ControlChinese
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Opus 4.6: Snapshot: claude-opus-4-6-high, Reasoning: high, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 1,488 published pairwise votes.GPT-5.6 Terra: Arena rating from 642 published pairwise votes.
Higher measured value1,351.953
95% confidence interval: ±17.967The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,313.871
95% confidence interval: ±24.287The 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 Vision, Style ControlCreative writing vision
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Opus 4.6: Snapshot: claude-opus-4-6, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 1,731 published pairwise votes.GPT-5.6 Terra: Arena rating from 914 published pairwise votes.
Higher measured value1,297.75
95% confidence interval: ±15.034The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,255.12
95% confidence interval: ±20.975The 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 Vision, Style ControlCreative writing vision
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Opus 4.6: Snapshot: claude-opus-4-6-high, Reasoning: high, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 1,405 published pairwise votes.GPT-5.6 Terra: Arena rating from 914 published pairwise votes.
Higher measured value1,309.447
95% confidence interval: ±16.801The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,255.12
95% confidence interval: ±20.975The 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 Vision, Style ControlDiagram
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Opus 4.6: Snapshot: claude-opus-4-6, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 7,260 published pairwise votes.GPT-5.6 Terra: Arena rating from 2,891 published pairwise votes.
Higher measured value1,331.371
95% confidence interval: ±8.825The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,300.563
95% confidence interval: ±11.707The 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 Vision, Style ControlDiagram
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Opus 4.6: Snapshot: claude-opus-4-6-high, Reasoning: high, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 5,999 published pairwise votes.GPT-5.6 Terra: Arena rating from 2,891 published pairwise votes.
Higher measured value1,324.675
95% confidence interval: ±9.238The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,300.563
95% confidence interval: ±11.707The 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 Vision, Style ControlEnglish
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Opus 4.6: Snapshot: claude-opus-4-6, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 11,327 published pairwise votes.GPT-5.6 Terra: Arena rating from 3,785 published pairwise votes.
Higher measured value1,296.058
95% confidence interval: ±8.463The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,270.126
95% confidence interval: ±11.03The 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 Vision, Style ControlEnglish
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Opus 4.6: Snapshot: claude-opus-4-6-high, Reasoning: high, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 9,251 published pairwise votes.GPT-5.6 Terra: Arena rating from 3,785 published pairwise votes.
Higher measured value1,301.128
95% confidence interval: ±8.811The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,270.126
95% confidence interval: ±11.03The 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 Vision, Style ControlHomework
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Opus 4.6: Snapshot: claude-opus-4-6, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 3,708 published pairwise votes.GPT-5.6 Terra: Arena rating from 1,172 published pairwise votes.
1,311.92
95% confidence interval: ±10.944The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
Higher measured value1,326.623
95% confidence interval: ±17.485The 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 Vision, Style ControlHomework
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Opus 4.6: Snapshot: claude-opus-4-6-high, Reasoning: high, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 2,849 published pairwise votes.GPT-5.6 Terra: Arena rating from 1,172 published pairwise votes.
1,324.17
95% confidence interval: ±12.006The 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 value1,326.623
95% confidence interval: ±17.485The 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 Vision, Style ControlHumor
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Opus 4.6: Snapshot: claude-opus-4-6, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 1,025 published pairwise votes.GPT-5.6 Terra: Arena rating from 502 published pairwise votes.
Higher measured value1,294.209
95% confidence interval: ±19.436The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,237.903
95% confidence interval: ±27.081The 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 Vision, Style ControlHumor
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Opus 4.6: Snapshot: claude-opus-4-6-high, Reasoning: high, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 869 published pairwise votes.GPT-5.6 Terra: Arena rating from 502 published pairwise votes.
Higher measured value1,296.35
95% confidence interval: ±21.068The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,237.903
95% confidence interval: ±27.081The 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 Vision, Style ControlOcr
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Opus 4.6: Snapshot: claude-opus-4-6, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 19,427 published pairwise votes.GPT-5.6 Terra: Arena rating from 7,536 published pairwise votes.
Higher measured value1,312.247
95% confidence interval: ±6.33The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,277.413
95% confidence interval: ±7.907The 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 Vision, Style ControlOcr
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterClaude Opus 4.6: Snapshot: claude-opus-4-6-high, Reasoning: high, Harness: Arena Vision, style control
GPT-5.6 Terra: Snapshot: gpt-5.6-terra-xhigh, Reasoning: xhigh, Harness: Arena Vision, style controlClaude Opus 4.6: Arena rating from 15,929 published pairwise votes.GPT-5.6 Terra: Arena rating from 7,536 published pairwise votes.
Higher measured value1,313.477
95% confidence interval: ±6.575The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,277.413
95% confidence interval: ±7.907The 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

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: 43b8c5c46830

Claude Opus 4.6

Claude Opus 4.6: SVG short test
API reports 0 reasoning tokensTest run: Oct 14, 2026Provider: anthropicAPI model: anthropic/claude-opus-4.6766 output tokens

GPT-5.6 Terra

Unavailable in this round

Earlier tests

Pelican on a bicycle

A community classic for free-form SVG drawing.

Claude Opus 4.6

Claude Opus 4.6: Pelican on a bicycle
A community classic for free-form SVG drawing.

Reasoning (requested): high

Current attempt
Run date
Oct 1, 2026
API provider
openrouter
Requested API model
anthropic/claude-opus-4.6
API model reported in response
anthropic/claude-opus-4.6
Reasoning setting
high
Pinned endpoint
Anthropic | anthropic/claude-4.6-opus-20260205
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
85.1 s

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

GPT-5.6 Terra

GPT-5.6 Terra: Pelican on a bicycle
A community classic for free-form SVG drawing.

Reasoning (requested): high

Current attempt
Run date
Sep 7, 2026
API provider
openrouter
Requested API model
openai/gpt-5.6-terra
API model reported in response
openai/gpt-5.6-terra
Reasoning setting
high
Pinned endpoint
OpenAI | openai/gpt-5.6-terra-20260709
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
64.6 s

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

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

Related LLM comparisons

Compare Claude Opus 4.6 and GPT-5.6 Terra with other leading models using the same data and benchmark logic.

View the complete LLM comparison