Claude Sonnet 4.6 vs GPT-5.4
Wins 0 of 8 areas
—
Wins 8 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Math · Long documents · Following instructions
GPT-5.4 is the stronger all-rounder.
Scores updated · 61 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.
- ReasoningHard problems that need careful thinking05GPT-5.45 of 5 tests
- AgentsCarrying out multi-step tasks on its own15GPT-5.45 of 7 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video03GPT-5.43 of 3 tests
- CodingWriting and fixing software24GPT-5.44 of 7 tests · 1 tie
- FactsGetting facts right instead of making them up13GPT-5.43 of 4 tests
- MathCompetition and research-level math02GPT-5.42 of 2 tests
- Following instructionsDoing exactly what it is asked02GPT-5.42 of 2 tests
- Long documentsFinding answers in very long texts01GPT-5.41 of 1 test
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 3% 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+40.2points ahead
- Research-level physics problemsCritPt+20.3points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+17.3points ahead
Where Claude Sonnet 4.6 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+43.3points ahead
- Finds the root cause of IT system incidentsAA IT-Bench SRE+5.3points ahead
- Programming problems, refreshed regularlyLiveBench · Coding+1.8points ahead
Every test, side by side
All 61 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5.4
- LiveBench · Agentic CodingGPT-5.4 by 11.242.653.8+11.2
- Terminal-Bench 2.1GPT-5.4 by 7.171.278.3+7.1
- SciCodeGPT-5.4 by 6.550.156.6+6.5
- LMArena · WebDevClaude Sonnet 4.6 by 57 rating points15221465+57 rating
- Terminal-Bench HardGPT-5.4 by 4.65357.6+4.6
- LiveBench · CodingClaude Sonnet 4.6 by 1.879.377.5+1.8
- SWE-bench Protie58.157.7tie
AgentsGPT-5.4
- BrowseCompGPT-5.4 by 874.782.7+8
- MCP AtlasGPT-5.4 by 5.961.367.2+5.9
- AA ApexAgentsGPT-5.4 by 5.32833.3+5.3
- AA IT-Bench SREClaude Sonnet 4.6 by 5.339.834.5+5.3
- τ-Bench V3 · BankingGPT-5.4 by 5.234.439.6+5.2
- OSWorld-VerifiedGPT-5.4 by 2.572.575+2.5
- GDPValtie36.737.4tie
ReasoningGPT-5.4
- CritPtGPT-5.4 by 20.33.123.4+20.3
- ARC-AGI-2GPT-5.4 by 1558.373.3+15
- Humanity's Last ExamGPT-5.4 by 10.133.643.7+10.1
- GPQA DiamondGPT-5.4 by 4.587.592+4.5
- LiveBench · ReasoningGPT-5.4 by 3.384.888.1+3.3
FactsGPT-5.4
- AA-Omniscience · Non-hallucinationClaude Sonnet 4.6 by 43.351.68.3+43.3
- SimpleQA VerifiedGPT-5.4 by 12.332.845.1+12.3
- AA-Omniscience · AccuracyGPT-5.4 by 1040.950.9+10
- Vectara HHEM hallucination ratelower is betterGPT-5.4 by 3.610.67+3.6
Images and chartsGPT-5.4
- CharXiv (RQ)GPT-5.4 by 11.271.682.8+11.2
- MMMU-ProGPT-5.4 by 5.173.378.4+5.1
- LMArena · VisionGPT-5.4 by 19 rating points12831302+19 rating
MathGPT-5.4
- USAMO 2026GPT-5.4 by 40.25595.2+40.2
- LiveBench · MathematicsGPT-5.4 by 7.28794.2+7.2
Following instructionsGPT-5.4
- IFBenchGPT-5.4 by 17.356.673.9+17.3
- LiveBench · Instruction FollowingGPT-5.4 by 763.270.2+7
Other results30 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.
- DeepSWE 1.1GPT-5.4 by 223052+22
- frontiermath_tier_4_v1GPT-5.4 by 18.88.327.1+18.8
- Terminal-Bench 2.0GPT-5.4 by 1659.175.1+16
- ToolathlonGPT-5.4 by 13.64154.6+13.6
- τ²-Bench Telecom (AA run)GPT-5.4 by 11.475.787.1+11.4
- AA Agentic IndexGPT-5.4 by 11.133.144.2+11.1
- AA IntelligenceGPT-5.4 by 8.930.139+8.9
- AgentWorldBench - SearchGPT-5.4 by 8.528.837.3+8.5
- Artificial Analysis Coding IndexGPT-5.4 by 8.16371.1+8.1
- ARC-AGI-1GPT-5.4 by 7.78693.7+7.7
- Finance AgentClaude Sonnet 4.6 by 7.363.356+7.3
- livebench_languageGPT-5.4 by 6.576.182.6+6.5
- AA-OmniscienceClaude Sonnet 4.6 by 6.412.25.8+6.4
- AgentWorldBench - OSGPT-5.4 by 5.463.268.6+5.4
- HLE (with tools)GPT-5.4 by 5.246.852+5.2
- Legal Agent BenchmarkClaude Sonnet 4.6 by 55.40.4+5
- LiveBenchGPT-5.4 by 4.875.580.3+4.8
- vectara_factual_consistencyGPT-5.4 by 3.689.493+3.6
- AgentWorldBench - TerminalClaude Sonnet 4.6 by 3.35753.7+3.3
- vectara_avg_summary_lengthClaude Sonnet 4.6 by 33 rating points114.781.7+33 rating
- AgentWorldBench - OverallGPT-5.4 by 2.35658.3+2.3
- OfficeQA ProClaude Sonnet 4.6 by 2.353.451.1+2.3
- AgentWorldBench - AndroidGPT-5.4 by 25860+2
- AgentWorldBench - SWEGPT-5.4 by 1.864.566.3+1.8
- livebench_data_analysisGPT-5.4 by 1.37879.3+1.3
- AgentWorldBench - WebGPT-5.4 by 150.851.8+1
- OfficeQAtie68.768.1tie
- GDPval-AA (Elo)tie16761674tie
- AgentWorldBench - MCPtie7070.1tie
- vectara_answer_ratetie99.999.9tie
Questions people ask
Which is better, Claude Sonnet 4.6 or GPT-5.4?
GPT-5.4 wins all eight areas we test: coding, agents, reasoning, facts, images and charts, math, long documents and following instructions. Claude Sonnet 4.6 wins none.
Which is better for coding?
GPT-5.4. It wins 4 of the 7 coding tests both models report; Claude Sonnet 4.6 wins 2, and 1 is a tie.
Which is cheaper?
Claude Sonnet 4.6 costs $3.00 per million input tokens and $15.00 per million output tokens; GPT-5.4 costs $2.50 and $15.00. That makes GPT-5.4 about 3% cheaper for the same work.
How do you compare the two?
We use the 61 benchmark tests both models have published scores on. The verdict counts the 31 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 30 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.