Claude Opus 4.5 vs GPT-5.2
Wins 1 of 8 areas
Facts
Wins 5 of 8 areas
Agents · Reasoning · Images and charts · Math · Following instructions
GPT-5.2 is the stronger all-rounder.Claude Opus 4.5 is better at facts.
Scores updated · 78 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 math06GPT-5.26 of 6 tests
- ReasoningHard problems that need careful thinking05GPT-5.25 of 5 tests
- Images and chartsUnderstanding pictures, charts and video12GPT-5.22 of 4 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own12GPT-5.22 of 3 tests
- Following instructionsDoing exactly what it is asked01GPT-5.21 of 2 tests · 1 tie
- FactsGetting facts right instead of making them up30Claude Opus 4.53 of 4 tests · 1 tie
- CodingWriting and fixing software33Even3 each · 2 ties
- Long documentsFinding answers in very long texts11Even1 each
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 48% 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
- Fixes real GitHub issues in many programming languagesSWE-bench Multilingual+59.7points ahead
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+33points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+28.8points ahead
Where Claude Opus 4.5 pulls ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+22.1points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+20.2points ahead
- Questions about very long textsLongBench v2+9.9points ahead
Every test, side by side
All 78 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingEven
- SWE-bench MultilingualGPT-5.2 by 59.7766.7+59.7
- SWE-bench ProClaude Opus 4.5 by 22.15229.9+22.1
- LiveBench · Agentic CodingGPT-5.2 by 10.639.750.3+10.6
- LMArena · WebDevClaude Opus 4.5 by 53 rating points14691416+53 rating
- LiveBench · CodingClaude Opus 4.5 by 3.679.776.1+3.6
- SciCodeGPT-5.2 by 2.649.552.1+2.6
- SWE-bench Verifiedtie80.980tie
- Terminal-Bench Hardtie4747tie
AgentsGPT-5.2
- BrowseCompGPT-5.2 by 28.83765.8+28.8
- MCP AtlasClaude Opus 4.5 by 1.762.360.6+1.7
- GDPValGPT-5.2 by 147.348.3+1
ReasoningGPT-5.2
- ARC-AGI-2GPT-5.2 by 15.337.652.9+15.3
- Humanity's Last ExamGPT-5.2 by 7.630.137.7+7.6
- CritPtGPT-5.2 by 74.611.6+7
- GPQA DiamondGPT-5.2 by 3.786.690.3+3.7
- LiveBench · ReasoningGPT-5.2 by 3.180.183.2+3.1
FactsClaude Opus 4.5
- AA-Omniscience · Non-hallucinationClaude Opus 4.5 by 20.23918.8+20.2
- SimpleQA VerifiedClaude Opus 4.5 by 8.645.737.1+8.6
- AA-Omniscience · AccuracyClaude Opus 4.5 by 2.346.644.3+2.3
- Vectara HHEM hallucination ratelower is bettertie10.910.8tie
Images and chartsGPT-5.2
- CharXiv (RQ)GPT-5.2 by 14.967.282.1+14.9
- OCRBenchClaude Opus 4.5 by 5.886.580.7+5.8
- MathVistaGPT-5.2 by 2.680.282.8+2.6
- MMMU-Protie7474.6tie
MathGPT-5.2
- FrontierMath Tiers 1-3 (v2)GPT-5.2 by 3334.467.4+33
- FrontierMath Tier 4GPT-5.2 by 26.84.931.7+26.8
- HMMT Feb. 2026GPT-5.2 by 11.785.397+11.7
- AIME 2026GPT-5.2 by 3.295.198.3+3.2
- LiveBench · MathematicsGPT-5.2 by 2.890.493.2+2.8
- IMOAnswerBenchGPT-5.2 by 2.38486.3+2.3
Long documentsEven
- LongBench v2Claude Opus 4.5 by 9.964.454.5+9.9
- AA-LCRGPT-5.2 by 5.477.382.7+5.4
Following instructionsGPT-5.2
- IFBenchGPT-5.2 by 17.45875.4+17.4
- LiveBench · Instruction Followingtie62.561.8tie
Other results44 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-OmniscienceClaude Opus 4.5 by 14.914-0.9+14.9
- frontiermath_tier_4_v1GPT-5.2 by 14.64.218.8+14.6
- SimpleVQAClaude Opus 4.5 by 13.969.755.8+13.9
- τ²-BenchClaude Opus 4.5 by 12.798.285.5+12.7
- BrowseComp (context management)GPT-5.2 by 12.257.870+12.2
- PaperBenchClaude Opus 4.5 by 9.272.963.7+9.2
- WorldVQAClaude Opus 4.5 by 8.836.828+8.8
- AIME 2025GPT-5.2 by 7.292.8100+7.2
- vectara_avg_summary_lengthGPT-5.2 by 71.8 rating points114.5186.3+71.8 rating
- HMMT Feb. 2025GPT-5.2 by 7.192.9100+7.1
- InfoVQA (val)GPT-5.2 by 7.176.984+7.1
- AIME25 no toolsGPT-5.2 by 79198+7
- τ²-Bench (Retail)Claude Opus 4.5 by 6.988.982+6.9
- ZeroBenchGPT-5.2 by 639+6
- MATH-VisionGPT-5.2 by 5.977.183+5.9
- swe_bench_bashClaude Opus 4.5 by 5.474.469+5.4
- Terminal-Bench 2.0Claude Opus 4.5 by 5.359.354+5.3
- metr_re_benchClaude Opus 4.5 by 54035+5
- DeepSearchQA (F1)Claude Opus 4.5 by 4.876.171.3+4.8
- τ²-Bench Telecom (AA run)Claude Opus 4.5 by 4.789.584.8+4.7
- MotionBenchGPT-5.2 by 4.560.364.8+4.5
- ARC-AGI-1GPT-5.2 by 4.28286.2+4.2
- livebench_data_analysisGPT-5.2 by 3.874.478.2+3.8
- metr_hcastGPT-5.2 by 3.673.477+3.6
- MMVUGPT-5.2 by 3.577.380.8+3.5
- MMLU-ProClaude Opus 4.5 by 2.889.586.7+2.8
- ToolathlonGPT-5.2 by 2.843.546.3+2.8
- HMMT Nov. 2025GPT-5.2 by 2.593.395.8+2.5
- HLE (with tools)GPT-5.2 by 2.343.245.5+2.3
- LiveCodeBenchGPT-5.2 by 28789+2
- OmniDocBench 1.5Claude Opus 4.5 by 287.785.7+2
- ZeroBench (w/ tools)Claude Opus 4.5 by 297+2
- livebench_languageClaude Opus 4.5 by 1.581.379.8+1.5
- VideoMMMUGPT-5.2 by 1.584.485.9+1.5
- AA IntelligenceGPT-5.2 by 1.329.130.4+1.3
- vectara_answer_rateGPT-5.2 by 1.398.7100+1.3
- LiveBenchClaude Opus 4.5 by 1.27674.8+1.2
- MMMLUClaude Opus 4.5 by 1.290.889.6+1.2
- SWE-PerfClaude Opus 4.5 by 1.14.73.6+1.1
- Artificial Analysis Coding Indextie47.848.7tie
- AA Agentic Indextie59.660.2tie
- SWT-benchtie80.280.7tie
- metr_swaatie99.599.6tie
- vectara_factual_consistencytie89.189.2tie
Questions people ask
Which is better, Claude Opus 4.5 or GPT-5.2?
GPT-5.2 wins five of the eight areas we test: agents, reasoning, images and charts, math and following instructions. Claude Opus 4.5 wins facts. They are level on coding and long documents.
Which is better for coding?
Neither. They win 3 coding tests each of the 8 both models report, and 2 are ties.
Which is cheaper?
Claude Opus 4.5 costs $5.00 per million input tokens and $25.00 per million output tokens; GPT-5.2 costs $1.75 and $14.00. That makes GPT-5.2 about 48% cheaper for the same work.
How do you compare the two?
We use the 78 benchmark tests both models have published scores on. The verdict counts the 34 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 44 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.