Claude Opus 4.8 vs GPT-5.4
Wins 6 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Math
Wins 1 of 8 areas
Long documents
Claude Opus 4.8 is the stronger all-rounder.GPT-5.4 is cheaper and better at long documents.
Scores updated · 69 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.
- AgentsCarrying out multi-step tasks on its own51Claude Opus 4.85 of 6 tests
- CodingWriting and fixing software52Claude Opus 4.85 of 8 tests · 1 tie
- MathCompetition and research-level math32Claude Opus 4.83 of 7 tests · 2 ties
- ReasoningHard problems that need careful thinking32Claude Opus 4.83 of 6 tests · 1 tie
- FactsGetting facts right instead of making them up21Claude Opus 4.82 of 3 tests
- Images and chartsUnderstanding pictures, charts and video10Claude Opus 4.81 of 2 tests · 1 tie
- Long documentsFinding answers in very long texts01GPT-5.41 of 1 test
- Following instructionsDoing exactly what it is asked11Even1 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 Claude Opus 4.8 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+52.4points ahead
- Multi-step tasks using many real software toolsMCP Atlas+15points ahead
- Fixes real GitHub issues in many programming languagesSWE-bench Multilingual+12.7points ahead
Where GPT-5.4 pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+11.7points ahead
- Olympiad math problems with short answersIMOAnswerBench+7.9points ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+5.4points ahead
Every test, side by side
All 69 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Opus 4.8
- SWE-bench MultilingualClaude Opus 4.8 by 12.784.471.7+12.7
- SWE-bench ProClaude Opus 4.8 by 11.569.257.7+11.5
- LMArena · WebDevClaude Opus 4.8 by 91 rating points15561465+91 rating
- Terminal-Bench 2.1Claude Opus 4.8 by 6.384.678.3+6.3
- LiveBench · CodingClaude Opus 4.8 by 4.381.877.5+4.3
- LiveBench · Agentic CodingGPT-5.4 by 3.350.553.8+3.3
- SciCodeGPT-5.4 by 2.254.456.6+2.2
- Terminal-Bench Hardtie58.357.6tie
AgentsClaude Opus 4.8
- MCP AtlasClaude Opus 4.8 by 1582.267.2+15
- GDPValClaude Opus 4.8 by 10.447.837.4+10.4
- OSWorld-VerifiedClaude Opus 4.8 by 8.483.475+8.4
- AA ApexAgentsClaude Opus 4.8 by 6.139.433.3+6.1
- τ-Bench V3 · BankingGPT-5.4 by 5.434.239.6+5.4
- BrowseCompClaude Opus 4.8 by 1.684.382.7+1.6
ReasoningClaude Opus 4.8
- Humanity's Last ExamClaude Opus 4.8 by 548.743.7+5
- CritPtGPT-5.4 by 2.520.923.4+2.5
- ARC-AGI-3Claude Opus 4.8 by 1.31.50.2+1.3
- ARC-AGI-2GPT-5.4 by 1.272.173.3+1.2
- LiveBench · ReasoningClaude Opus 4.8 by 1.189.288.1+1.1
- GPQA Diamondtie9292tie
FactsClaude Opus 4.8
- AA-Omniscience · Non-hallucinationClaude Opus 4.8 by 52.460.78.3+52.4
- SimpleQA VerifiedClaude Opus 4.8 by 7.95345.1+7.9
- AA-Omniscience · AccuracyGPT-5.4 by 2.148.850.9+2.1
Images and chartsClaude Opus 4.8
- CharXiv (RQ)Claude Opus 4.8 by 7.189.982.8+7.1
- LMArena · Visiontie12951302tie
MathClaude Opus 4.8
- IMOAnswerBenchGPT-5.4 by 7.983.591.4+7.9
- FrontierMath Tier 4Claude Opus 4.8 by 7.156.149+7.1
- HMMT Feb. 2026GPT-5.4 by 2.295.597.7+2.2
- USAMO 2026Claude Opus 4.8 by 1.596.795.2+1.5
- FrontierMath Tiers 1-3 (v2)Claude Opus 4.8 by 1.48078.6+1.4
- AIME 2026tie10099.2tie
- LiveBench · Mathematicstie94.394.2tie
Following instructionsEven
- IFBenchGPT-5.4 by 11.762.273.9+11.7
- LiveBench · Instruction FollowingClaude Opus 4.8 by 1.87270.2+1.8
Other results34 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.
- NL2RepoClaude Opus 4.8 by 28.469.741.3+28.4
- HiL-Bench (Tools-allowed)Claude Opus 4.8 by 25.635.39.7+25.6
- AA-OmniscienceClaude Opus 4.8 by 2328.85.8+23
- GDPval-AA (Elo)Claude Opus 4.8 by 216 rating points18901674+216 rating
- AgentWorldBench - MCPGPT-5.4 by 15.254.970.1+15.2
- OfficeQA ProClaude Opus 4.8 by 15.166.251.1+15.1
- DeepSearchQA (F1)Claude Opus 4.8 by 14.593.178.6+14.5
- livebench_data_analysisGPT-5.4 by 13.36679.3+13.3
- CyberGymClaude Opus 4.8 by 12.578.866.3+12.5
- matharena_visual_math_overallGPT-5.4 by 10.981.692.5+10.9
- Legal Agent BenchmarkClaude Opus 4.8 by 9.6100.4+9.6
- OfficeQAClaude Opus 4.8 by 9.577.668.1+9.5
- τ²-Bench Telecom (AA run)Claude Opus 4.8 by 7.394.487.1+7.3
- DeepSWE 1.1Claude Opus 4.8 by 75952+7
- HLE (with tools)Claude Opus 4.8 by 5.957.952+5.9
- AgentWorldBench - TerminalClaude Opus 4.8 by 5.559.253.7+5.5
- Tool-DecathlonClaude Opus 4.8 by 5.359.954.6+5.3
- ToolathlonClaude Opus 4.8 by 5.359.954.6+5.3
- frontiermath_tier_4_v1Claude Opus 4.8 by 4.231.327.1+4.2
- Artificial Analysis Coding IndexClaude Opus 4.8 by 3.274.371.1+3.2
- LiveBenchGPT-5.4 by 3.177.280.3+3.1
- AgentWorldBench - WebClaude Opus 4.8 by 2.954.751.8+2.9
- livebench_languageGPT-5.4 by 2.979.782.6+2.9
- AA IntelligenceClaude Opus 4.8 by 2.841.839+2.8
- AgentWorldBench - SearchGPT-5.4 by 2.235.137.3+2.2
- AgentWorldBench - SWEGPT-5.4 by 2.264.166.3+2.2
- Finance AgentGPT-5.4 by 2.153.956+2.1
- AgentWorldBench - OSGPT-5.4 by 266.668.6+2
- AgentWorldBench - OverallGPT-5.4 by 1.756.658.3+1.7
- AA Agentic IndexGPT-5.4 by 1.642.644.2+1.6
- AgentWorldBench - AndroidClaude Opus 4.8 by 1.561.560+1.5
- ARC-AGI-1GPT-5.4 by 1.292.593.7+1.2
- HMMT Nov. 2025tie96.595.8tie
- Terminal-Bench 2.0tie74.675.1tie
Questions people ask
Which is better, Claude Opus 4.8 or GPT-5.4?
Claude Opus 4.8 wins six of the eight areas we test: coding, agents, reasoning, facts, images and charts and math. GPT-5.4 wins long documents, and costs 42% less. They are level on following instructions.
Which is better for coding?
Claude Opus 4.8. It wins 5 of the 8 coding tests both models report; GPT-5.4 wins 2, and 1 is a tie.
Which is cheaper?
Claude Opus 4.8 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 69 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 34 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.