Claude Sonnet 4.6 vs GPT-5.2
Wins 2 of 8 areas
Coding · Agents
Wins 4 of 8 areas
Reasoning · Facts · Images and charts · Math
GPT-5.2 wins more areas, narrowly.Claude Sonnet 4.6 is better at coding.
Scores updated · 49 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 thinking23GPT-5.23 of 5 tests
- FactsGetting facts right instead of making them up12GPT-5.22 of 4 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video12GPT-5.22 of 3 tests
- MathCompetition and research-level math01GPT-5.21 of 1 test
- CodingWriting and fixing software42Claude Sonnet 4.64 of 7 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own31Claude Sonnet 4.63 of 5 tests · 1 tie
- Long documentsFinding answers in very long texts11Even1 each
- Following instructionsDoing exactly what it is asked11Even1 each
Math 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-5.2 costs 13% 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-5.2 pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+18.8points ahead
- Real work tasks from 44 professionsGDPVal+11.6points ahead
- Reasoning about charts from research papersCharXiv (RQ)+10.5points ahead
Where Claude Sonnet 4.6 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+32.8points ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+28.2points ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+23.3points ahead
Every test, side by side
All 49 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Sonnet 4.6
- SWE-bench ProClaude Sonnet 4.6 by 28.258.129.9+28.2
- LMArena · WebDevClaude Sonnet 4.6 by 106 rating points15211415+106 rating
- LiveBench · Agentic CodingGPT-5.2 by 7.742.650.3+7.7
- Terminal-Bench HardClaude Sonnet 4.6 by 65347+6
- LiveBench · CodingClaude Sonnet 4.6 by 3.279.376.1+3.2
- SciCodeGPT-5.2 by 250.152.1+2
- SWE-bench Verifiedtie79.680tie
AgentsClaude Sonnet 4.6
- τ-Bench V3 · BankingClaude Sonnet 4.6 by 23.334.411.1+23.3
- GDPValGPT-5.2 by 11.636.748.3+11.6
- BrowseCompClaude Sonnet 4.6 by 8.974.765.8+8.9
- AA ApexAgentsClaude Sonnet 4.6 by 52823+5
- MCP Atlastie61.360.6tie
ReasoningGPT-5.2
- CritPtGPT-5.2 by 8.53.111.6+8.5
- ARC-AGI-2Claude Sonnet 4.6 by 5.458.352.9+5.4
- Humanity's Last ExamGPT-5.2 by 4.133.637.7+4.1
- GPQA DiamondGPT-5.2 by 2.887.590.3+2.8
- LiveBench · ReasoningClaude Sonnet 4.6 by 1.684.883.2+1.6
FactsGPT-5.2
- AA-Omniscience · Non-hallucinationClaude Sonnet 4.6 by 32.851.618.8+32.8
- SimpleQA VerifiedGPT-5.2 by 4.332.837.1+4.3
- AA-Omniscience · AccuracyGPT-5.2 by 3.440.944.3+3.4
- Vectara HHEM hallucination ratelower is bettertie10.610.8tie
Images and chartsGPT-5.2
- CharXiv (RQ)GPT-5.2 by 10.571.682.1+10.5
- LMArena · VisionClaude Sonnet 4.6 by 37 rating points12831246+37 rating
- MMMU-ProGPT-5.2 by 1.373.374.6+1.3
Long documentsEven
- AA-LCRGPT-5.2 by 2.78082.7+2.7
- MRCR v2 8 needle 128k (average)Claude Sonnet 4.6 by 1.184.983.8+1.1
Following instructionsEven
- IFBenchGPT-5.2 by 18.856.675.4+18.8
- LiveBench · Instruction FollowingClaude Sonnet 4.6 by 1.463.261.8+1.4
Other results20 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.
- AA Agentic IndexGPT-5.2 by 27.133.160.2+27.1
- GDPval-AA (Elo)Claude Sonnet 4.6 by 214 rating points16761462+214 rating
- Artificial Analysis Coding IndexClaude Sonnet 4.6 by 14.36348.7+14.3
- AA-OmniscienceClaude Sonnet 4.6 by 13.112.2-0.9+13.1
- frontiermath_tier_4_v1GPT-5.2 by 10.58.318.8+10.5
- τ²-Bench (Retail)Claude Sonnet 4.6 by 9.791.782+9.7
- τ²-Bench Telecom (AA run)GPT-5.2 by 9.175.784.8+9.1
- vectara_avg_summary_lengthGPT-5.2 by 71.6 rating points114.7186.3+71.6 rating
- ToolathlonGPT-5.2 by 5.34146.3+5.3
- Terminal-Bench 2.0Claude Sonnet 4.6 by 5.159.154+5.1
- livebench_languageGPT-5.2 by 3.776.179.8+3.7
- GPQA (unspecified)GPT-5.2 by 2.589.992.4+2.5
- HLE (with tools)Claude Sonnet 4.6 by 1.346.845.5+1.3
- LiveBenchtie75.574.8tie
- AA Intelligencetie30.130.4tie
- MMMLUtie89.389.6tie
- ARC-AGI-1tie8686.2tie
- livebench_data_analysistie7878.2tie
- vectara_factual_consistencytie89.489.2tie
- vectara_answer_ratetie99.9100tie
Questions people ask
Which is better, Claude Sonnet 4.6 or GPT-5.2?
GPT-5.2 wins four of the eight areas we test: reasoning, facts, images and charts and math. Claude Sonnet 4.6 wins coding and agents. They are level on long documents and following instructions.
Which is better for coding?
Claude Sonnet 4.6. It wins 4 of the 7 coding tests both models report; GPT-5.2 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-5.2 costs $1.75 and $14.00. That makes GPT-5.2 about 13% cheaper for the same work.
How do you compare the two?
We use the 49 benchmark tests both models have published scores on. The verdict counts the 29 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 20 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.