Claude Opus 4.6 vs GPT-5.4
Wins 0 of 8 areas
—
Wins 6 of 8 areas
Coding · Reasoning · Images and charts · Math · Long documents · Following instructions
GPT-5.4 is the stronger all-rounder.
Scores updated · 88 tests both models report · How we compare
Where each one wins
Tests won in each of the eight areas we test. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- MathCompetition and research-level math07GPT-5.47 of 7 tests
- ReasoningHard problems that need careful thinking04GPT-5.44 of 6 tests · 2 ties
- CodingWriting and fixing software24GPT-5.44 of 7 tests · 1 tie
- Following instructionsDoing exactly what it is asked02GPT-5.42 of 2 tests
- Images and chartsUnderstanding pictures, charts and video12GPT-5.42 of 3 tests
- Long documentsFinding answers in very long texts01GPT-5.41 of 1 test
- AgentsCarrying out multi-step tasks on its own22Even2 each · 1 tie
- FactsGetting facts right instead of making them up22Even2 each
Long documents rests on a single test.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
GPT-5.4 costs 42% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where GPT-5.4 pulls ahead
- Proof problems from the US Math OlympiadUSAMO 2026+29points ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+22.2points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+20.8points ahead
Where Claude Opus 4.6 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+28.9points ahead
- Real work tasks from 44 professionsGDPVal+18.5points ahead
- Fixes real GitHub issues in many programming languagesSWE-bench Multilingual+6.1points ahead
Every test, side by side
All 88 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5.4
- Terminal-Bench HardGPT-5.4 by 11.446.257.6+11.4
- LMArena · WebDevClaude Opus 4.6 by 81 rating points15461465+81 rating
- SWE-bench MultilingualClaude Opus 4.6 by 6.177.871.7+6.1
- LiveBench · Agentic CodingGPT-5.4 by 4.84953.8+4.8
- SciCodeGPT-5.4 by 4.751.956.6+4.7
- SWE-bench ProGPT-5.4 by 4.353.457.7+4.3
- LiveBench · Codingtie78.277.5tie
AgentsEven
- GDPValClaude Opus 4.6 by 18.555.937.4+18.5
- MCP AtlasGPT-5.4 by 4.562.767.2+4.5
- OSWorld-VerifiedGPT-5.4 by 2.372.775+2.3
- BrowseCompClaude Opus 4.6 by 1.38482.7+1.3
- AA ApexAgentstie3333.3tie
ReasoningGPT-5.4
- CritPtGPT-5.4 by 10.812.623.4+10.8
- ARC-AGI-2GPT-5.4 by 4.568.873.3+4.5
- Humanity's Last ExamGPT-5.4 by 3.839.943.7+3.8
- GPQA DiamondGPT-5.4 by 2.489.692+2.4
- LiveBench · Reasoningtie88.788.1tie
- ARC-AGI-3tie0.50.2tie
FactsEven
- AA-Omniscience · Non-hallucinationClaude Opus 4.6 by 28.937.28.3+28.9
- Vectara HHEM hallucination ratelower is betterGPT-5.4 by 5.212.27+5.2
- AA-Omniscience · AccuracyGPT-5.4 by 3.94750.9+3.9
- SimpleQA VerifiedClaude Opus 4.6 by 1.94745.1+1.9
Images and chartsGPT-5.4
- CharXiv (RQ)GPT-5.4 by 5.477.482.8+5.4
- MMMU-ProGPT-5.4 by 375.478.4+3
- LMArena · VisionClaude Opus 4.6 by 14 rating points13161302+14 rating
MathGPT-5.4
- USAMO 2026GPT-5.4 by 2966.295.2+29
- FrontierMath Tier 4GPT-5.4 by 22.226.849+22.2
- IMOAnswerBenchGPT-5.4 by 16.175.391.4+16.1
- FrontierMath Tiers 1-3 (v2)GPT-5.4 by 12.66678.6+12.6
- LiveBench · MathematicsGPT-5.4 by 4.989.394.2+4.9
- AIME 2026GPT-5.4 by 2.596.799.2+2.5
- HMMT Feb. 2026GPT-5.4 by 1.596.297.7+1.5
Following instructionsGPT-5.4
- IFBenchGPT-5.4 by 20.853.173.9+20.8
- LiveBench · Instruction FollowingGPT-5.4 by 6.963.370.2+6.9
Other results53 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- baby_vision_with_pythonGPT-5.4 by 41.838.480.2+41.8
- BabyVisionGPT-5.4 by 34.914.849.7+34.9
- HiL-Bench (Tools-allowed)Claude Opus 4.6 by 28.638.39.7+28.6
- MATH-VisionGPT-5.4 by 26.565.592+26.5
- AA Agentic IndexClaude Opus 4.6 by 23.467.644.2+23.4
- Artificial Analysis Coding IndexGPT-5.4 by 2348.171.1+23
- matharena_visual_math_overallGPT-5.4 by 20.272.392.5+20.2
- Apex (Pass@1)GPT-5.4 by 19.634.554.1+19.6
- DeepSearchQA (accuracy)Claude Opus 4.6 by 16.980.663.7+16.9
- DeepSearchQA (F1)Claude Opus 4.6 by 12.791.378.6+12.7
- V* (w/ python)GPT-5.4 by 1286.498.4+12
- mathvision_with_pythonGPT-5.4 by 11.584.696.1+11.5
- Terminal-Bench 2.0GPT-5.4 by 9.765.475.1+9.7
- livebench_data_analysisGPT-5.4 by 9.469.979.3+9.4
- NL2RepoClaude Opus 4.6 by 8.549.841.3+8.5
- AgentWorldBench - SearchGPT-5.4 by 829.337.3+8
- AA-OmniscienceClaude Opus 4.6 by 7.913.75.8+7.9
- Apex Shortlist (Pass@1)Claude Opus 4.6 by 7.885.978.1+7.8
- CyberGymClaude Opus 4.6 by 7.573.866.3+7.5
- Tool-DecathlonGPT-5.4 by 7.447.254.6+7.4
- AA IntelligenceGPT-5.4 by 7.131.939+7.1
- MCPAtlas Public (Pass@1)Claude Opus 4.6 by 6.673.867.2+6.6
- OfficeQA ProClaude Opus 4.6 by 657.151.1+6
- MCPMarkGPT-5.4 by 5.856.762.5+5.8
- vectara_avg_summary_lengthClaude Opus 4.6 by 55.9 rating points137.681.7+55.9 rating
- GDPval-AA (Elo)GPT-5.4 by 55 rating points16191674+55 rating
- OfficeQAClaude Opus 4.6 by 5.473.568.1+5.4
- charxiv_rq_with_pythonGPT-5.4 by 5.384.790+5.3
- vectara_factual_consistencyGPT-5.4 by 5.287.893+5.2
- τ²-Bench Telecom (AA run)Claude Opus 4.6 by 592.187.1+5
- mmmu_pro_with_pythonGPT-5.4 by 4.877.382.1+4.8
- Finance AgentClaude Opus 4.6 by 4.760.756+4.7
- frontiermath_tier_4_v1GPT-5.4 by 4.222.927.1+4.2
- LiveBenchGPT-5.4 by 476.380.3+4
- AgentWorldBench - TerminalClaude Opus 4.6 by 3.857.553.7+3.8
- Legal Agent BenchmarkClaude Opus 4.6 by 3.84.20.4+3.8
- ToolathlonClaude Opus 4.6 by 2.256.854.6+2.2
- AgentWorldBench - AndroidClaude Opus 4.6 by 1.761.760+1.7
- AgentWorldBench - SWEGPT-5.4 by 1.764.566.3+1.7
- AgentWorldBench - OSClaude Opus 4.6 by 1.670.268.6+1.6
- MMLU-ProClaude Opus 4.6 by 1.689.187.5+1.6
- browsecomp_with_context_managerClaude Opus 4.6 by 1.38482.7+1.3
- HLE (with tools)Claude Opus 4.6 by 1.353.352+1.3
- BrowseComp (Agent Swarm)Claude Opus 4.6 by 183.782.7+1
- ARC-AGI-1tie9393.7tie
- livebench_languagetie83.382.6tie
- AgentWorldBench - Overalltie57.858.3tie
- HMMT Nov. 2025tie96.395.8tie
- τ³-Benchtie72.472.9tie
- AgentWorldBench - Webtie51.451.8tie
- Chinese SimpleQA (C-SimpleQA)tie76.476.8tie
- AgentWorldBench - MCPtie69.970.1tie
- vectara_answer_ratetie99.899.9tie
Questions people ask
Which is better, Claude Opus 4.6 or GPT-5.4?
GPT-5.4 wins six of the eight areas we test: coding, reasoning, images and charts, math, long documents and following instructions. Claude Opus 4.6 wins none. They are level on agents and facts.
Which is better for coding?
GPT-5.4. It wins 4 of the 7 coding tests both models report; Claude Opus 4.6 wins 2, and 1 is a tie.
Which is cheaper?
Claude Opus 4.6 costs $5.00 per million input tokens and $25.00 per million output tokens; GPT-5.4 costs $2.50 and $15.00. That makes GPT-5.4 about 42% cheaper for the same work.
How do you compare the two?
We use the 88 benchmark tests both models have published scores on. The verdict counts the 35 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 53 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.