Claude Sonnet 4.5 vs GPT-5.2
Wins 0 of 8 areas
—
Wins 7 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Math · Following instructions
GPT-5.2 is the stronger all-rounder.
Scores updated · 59 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.
- CodingWriting and fixing software14GPT-5.24 of 6 tests · 1 tie
- ReasoningHard problems that need careful thinking14GPT-5.24 of 5 tests
- Images and chartsUnderstanding pictures, charts and video03GPT-5.23 of 3 tests
- MathCompetition and research-level math03GPT-5.23 of 3 tests
- AgentsCarrying out multi-step tasks on its own13GPT-5.23 of 4 tests
- FactsGetting facts right instead of making them up13GPT-5.23 of 4 tests
- Following instructionsDoing exactly what it is asked01GPT-5.21 of 1 test
- Long documentsFinding answers in very long texts11Even1 each
Following instructions 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
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+43.5points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+41.7points ahead
- Abstract visual puzzles that people can solveARC-AGI-2+39.3points ahead
Where Claude Sonnet 4.5 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+32.1points ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+13.7points ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+13.4points ahead
Every test, side by side
All 59 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5.2
- SWE-bench ProClaude Sonnet 4.5 by 13.743.629.9+13.7
- Terminal-Bench HardGPT-5.2 by 11.435.647+11.4
- SciCodeGPT-5.2 by 6.445.752.1+6.4
- SWE-bench VerifiedGPT-5.2 by 2.877.280+2.8
- LMArena · WebDevGPT-5.2 by 22 rating points13941416+22 rating
- SWE-bench Multilingualtie6766.7tie
AgentsGPT-5.2
- BrowseCompGPT-5.2 by 41.724.165.8+41.7
- GDPValGPT-5.2 by 27.820.548.3+27.8
- MCP AtlasGPT-5.2 by 16.843.860.6+16.8
- τ-Bench V3 · BankingClaude Sonnet 4.5 by 13.424.511.1+13.4
ReasoningGPT-5.2
- ARC-AGI-2GPT-5.2 by 39.313.652.9+39.3
- Humanity's Last ExamGPT-5.2 by 19.917.837.7+19.9
- CritPtGPT-5.2 by 10.51.111.6+10.5
- SimpleBenchClaude Sonnet 4.5 by 8.554.345.8+8.5
- GPQA DiamondGPT-5.2 by 6.983.490.3+6.9
FactsGPT-5.2
- AA-Omniscience · Non-hallucinationClaude Sonnet 4.5 by 32.150.918.8+32.1
- AA-Omniscience · AccuracyGPT-5.2 by 11.432.944.3+11.4
- SimpleQA VerifiedGPT-5.2 by 6.430.737.1+6.4
- Vectara HHEM hallucination ratelower is betterGPT-5.2 by 1.21210.8+1.2
Images and chartsGPT-5.2
- CharXiv (RQ)GPT-5.2 by 14.967.282.1+14.9
- MMMU-ProGPT-5.2 by 5.968.774.6+5.9
- MathVistaGPT-5.2 by 379.882.8+3
MathGPT-5.2
- FrontierMath Tiers 1-3 (v2)GPT-5.2 by 43.523.967.4+43.5
- FrontierMath Tier 4GPT-5.2 by 29.32.431.7+29.3
- IMOAnswerBenchGPT-5.2 by 20.465.986.3+20.4
Long documentsEven
- AA-LCRGPT-5.2 by 10.472.382.7+10.4
- LongBench v2Claude Sonnet 4.5 by 7.361.854.5+7.3
Other results31 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.
- BrowseComp (context management)GPT-5.2 by 43.926.170+43.9
- AA Agentic IndexGPT-5.2 by 42.717.560.2+42.7
- HMMT Feb. 2025GPT-5.2 by 32.567.5100+32.5
- HMMT 2025GPT-5.2 by 24.874.699.4+24.8
- ARC-AGI-1GPT-5.2 by 22.563.786.2+22.5
- LiveCodeBenchGPT-5.2 by 187189+18
- frontiermath_tier_4_v1GPT-5.2 by 14.64.218.8+14.6
- HMMT Nov. 2025GPT-5.2 by 14.181.795.8+14.1
- HLE (with tools)GPT-5.2 by 13.53245.5+13.5
- AIME 2025GPT-5.2 by 1387100+13
- τ²-BenchClaude Sonnet 4.5 by 12.59885.5+12.5
- SWT-benchGPT-5.2 by 11.269.580.7+11.2
- matharena_visual_math_overallGPT-5.2 by 10.775.886.5+10.7
- AIME25 no toolsGPT-5.2 by 108898+10
- AA IntelligenceGPT-5.2 by 9.720.730.4+9.7
- GPQA (unspecified)GPT-5.2 by 983.492.4+9
- τ²-Bench Telecom (AA run)GPT-5.2 by 6.778.184.8+6.7
- vectara_avg_summary_lengthGPT-5.2 by 58.5 rating points127.8186.3+58.5 rating
- ToolathlonGPT-5.2 by 5.34146.3+5.3
- vectara_answer_rateGPT-5.2 by 4.495.6100+4.4
- τ²-Bench (Retail)Claude Sonnet 4.5 by 4.286.282+4.2
- Terminal-Bench 2.0GPT-5.2 by 45054+4
- Artificial Analysis Coding IndexClaude Sonnet 4.5 by 3.452.148.7+3.4
- swe_bench_bashClaude Sonnet 4.5 by 2.471.469+2.4
- SimpleVQAClaude Sonnet 4.5 by 1.857.655.8+1.8
- vectara_factual_consistencyGPT-5.2 by 1.28889.2+1.2
- AA-Omnisciencetie-0.1-0.9tie
- MMLU-Protie87.586.7tie
- SWE-Perftie33.6tie
- MMMLUtie89.189.6tie
- OmniDocBench 1.5tie85.885.7tie
Questions people ask
Which is better, Claude Sonnet 4.5 or GPT-5.2?
GPT-5.2 wins seven of the eight areas we test: coding, agents, reasoning, facts, images and charts, math and following instructions. Claude Sonnet 4.5 wins none. They are level on long documents.
Which is better for coding?
GPT-5.2. It wins 4 of the 6 coding tests both models report; Claude Sonnet 4.5 wins 1, and 1 is a tie.
Which is cheaper?
Claude Sonnet 4.5 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 59 benchmark tests both models have published scores on. The verdict counts the 28 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 31 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.