Compare published benchmark results from matching versions and cohorts, API pricing, and technical specifications. Any documented test setup differences remain visible.
shared result series
138
distinct benchmarks
20
latest retrieval
2026-09-04
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.
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.
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 GPT-5.6 Luna and GPT-5.4 mini
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.
GPT-5.6 Luna
GPT-5.4 mini
Coding: 100 / 55.7 (5 benchmark families)
Reasoning: 100 / 90.6 (1 benchmark family)
Knowledge: 100 / 98.3 (1 benchmark family)
Finance: 100 / 86.6 (5 benchmark families)
Legal: 100 / 17.1 (2 benchmark families)
Agentic tasks: 100 / 69.2 (1 benchmark family)
Multimodal: 100 / 93.2 (1 benchmark family)
Education: 87 / 100 (1 benchmark family)
21 matched rows from 19 benchmark families were considered. The radar shows 8 of 9 comparable categories.
GPT-5.6 LunaGPT-5.4 mini
Code MigrationOverall. Higher is better. No documented setup difference
GPT-5.6 LunaWinner44.55 %
GPT-5.4 mini12.936 %
CorpFin v2Overall. Higher is better. No documented setup difference
GPT-5.6 LunaWinner64.219 %
GPT-5.4 mini60.917 %
EMBOverall. Higher is better. No documented setup difference
GPT-5.6 LunaWinner67.116 %
GPT-5.4 mini45.426 %
Finance Agent (v2)Overall. Higher is better. No documented setup difference
GPT-5.6 LunaWinner55.044 %
GPT-5.4 mini45.36 %
GPQA DiamondOverall. Higher is better. No documented setup difference
GPT-5.6 LunaWinner91.666 %
GPT-5.4 mini83.08 %
Harvey's Legal Agent BenchmarkOverall · Task fully resolved. Higher is better. No documented setup difference
GPT-5.6 LunaWinner1.25 %
GPT-5.4 mini0 %
IOIOverall. Higher is better. No documented setup difference
GPT-5.6 LunaWinner72.917 %
GPT-5.4 mini6.417 %
Legal Research BenchOverall · All-pass. Higher is better. No documented setup difference
GPT-5.6 LunaWinner36.538 %
GPT-5.4 mini12.5 %
LMArena Text, Style ControlOverall. Higher is better. Test setup not documented as identical
GPT-5.6 LunaHigher measured value1,452.364
GPT-5.4 mini1,448.118
MMLU-ProOverall. Higher is better. No documented setup difference
GPT-5.6 LunaWinner86.036 %
GPT-5.4 mini84.554 %
MMMU-ProOverall. Higher is better. No documented setup difference
GPT-5.6 LunaWinner85.029 %
GPT-5.4 mini79.249 %
MortgageTaxOverall. Higher is better. No documented setup difference
GPT-5.6 LunaWinner67.29 %
GPT-5.4 mini63.514 %
Key to the marks:better valuebetter, but not conclusivetie
Benchmark scores for GPT-5.6 Luna and GPT-5.4 mini
Benchmark and task
GPT-5.6 Luna
GPT-5.4 mini
Source
Code MigrationOverallTest details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is betterGPT-5.4 mini: Snapshot: gpt-5.4-mini-2026-03-17
No documented difference in the test setupVersion: 1Metric: task resolution rateScoring: Higher is betterGPT-5.4 mini: Snapshot: gpt-5.4-mini-2026-03-17
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control GPT-5.4 mini: Snapshot: gpt-5.4-mini-high, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 24,433 anonymous pairwise votes.GPT-5.4 mini: Style-controlled Arena rating from 59,472 anonymous pairwise votes.
Higher measured value1,452.364
95% confidence interval: ±5.03The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,448.118
95% confidence interval: ±3.851The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
No documented difference in the test setupVersion: 2Metric: weighted index scoreScoring: Higher is betterGPT-5.4 mini: Snapshot: gpt-5.4-mini-2026-03-17
No documented difference in the test setupVersion: 1.2Metric: weighted index scoreScoring: Higher is betterGPT-5.4 mini: Snapshot: gpt-5.4-mini-2026-03-17
No documented difference in the test setupVersion: 1.1Metric: accuracyScoring: Higher is betterGPT-5.6 Luna: Harness: OpenHands GPT-5.4 mini: Snapshot: gpt-5.4-mini-2026-03-17, Harness: OpenHands
LMArena Text, Style ControlBusiness, management and financeTest details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control GPT-5.4 mini: Snapshot: gpt-5.4-mini-high, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 4,691 anonymous pairwise votes.GPT-5.4 mini: Style-controlled Arena rating from 12,113 anonymous pairwise votes.
Higher measured value1,461.709
95% confidence interval: ±9.371The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,459.508
95% confidence interval: ±6.653The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control GPT-5.4 mini: Snapshot: gpt-5.4-mini-high, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 1,691 anonymous pairwise votes.GPT-5.4 mini: Style-controlled Arena rating from 3,298 anonymous pairwise votes.
1,474.185
95% confidence interval: ±15.484The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,477.451
95% confidence interval: ±11.125The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control GPT-5.4 mini: Snapshot: gpt-5.4-mini-high, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 6,843 anonymous pairwise votes.GPT-5.4 mini: Style-controlled Arena rating from 16,204 anonymous pairwise votes.
Higher measured value1,497.069
95% confidence interval: ±7.965The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,495.819
95% confidence interval: ±6.028The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Key to the marks:better valuebetter, but not conclusivetie
Our model tests
Community tests you can inspect
Compare how the models respond to the same prompt. Each test shows the first attempt, with no subsequent fixes to the generated code. These results do not contribute to an overall score.
Pelican on a bicycle
A community classic for free-form SVG drawing.
GPT-5.6 Luna
A community classic for free-form SVG drawing.First attempt
Run date
7 Sept 2026
API provider
openrouter
Requested API model
openai/gpt-5.6-luna
API model reported in response
openai/gpt-5.6-luna
Pinned endpoint
OpenAI | openai/gpt-5.6-luna-20260709
Reasoning setting
high
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
49.9 s
Visually reviewed. This individual test does not establish overall model quality.
GPT-5.4 mini
The response reached the token limit. No complete recording is available.
First attempt
Run date
7 Sept 2026
API provider
openrouter
Requested API model
openai/gpt-5.4-mini
API model reported in response
openai/gpt-5.4-mini
Pinned endpoint
OpenAI | openai/gpt-5.4-mini-20260317
Reasoning setting
high
Temperature
Provider default
Token limit including reasoning
8,192 tokens
Test protocol
community-visual-tests-v1
Technical checks
Failed
Response time
67 s
Not yet visually reviewed. A successful recording does not confirm correct physics or full compliance with the prompt.
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 GPT-5.6 Luna and GPT-5.4 mini with other leading models using the same data and benchmark logic.
