Claude Opus 4.7 vs GPT-5.2
Wins 5 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts
Wins 2 of 8 areas
Math · Long documents
Claude Opus 4.7 is the stronger all-rounder.GPT-5.2 is cheaper and better at long documents.
Scores updated · 58 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 software70Claude Opus 4.77 of 8 tests · 1 tie
- ReasoningHard problems that need careful thinking50Claude Opus 4.75 of 6 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video50Claude Opus 4.75 of 5 tests
- AgentsCarrying out multi-step tasks on its own41Claude Opus 4.74 of 5 tests
- FactsGetting facts right instead of making them up31Claude Opus 4.73 of 4 tests
- Long documentsFinding answers in very long texts02GPT-5.22 of 2 tests
- MathCompetition and research-level math12GPT-5.22 of 5 tests · 2 ties
- Following instructionsDoing exactly what it is asked11Even1 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 Claude Opus 4.7 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+38.9points ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+34.4points ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+23.5points ahead
Where GPT-5.2 pulls ahead
- Finds one of eight look-alike replies in a long chatMRCR v2 8 needle 128k (average)+24.5points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+16.8points ahead
- Real work tasks from 44 professionsGDPVal+5.5points ahead
Every test, side by side
All 58 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Opus 4.7
- SWE-bench ProClaude Opus 4.7 by 34.464.329.9+34.4
- LMArena · WebDevClaude Opus 4.7 by 142 rating points15571415+142 rating
- SWE-bench MultilingualClaude Opus 4.7 by 13.880.566.7+13.8
- SWE-bench VerifiedClaude Opus 4.7 by 7.687.680+7.6
- LiveBench · CodingClaude Opus 4.7 by 682.176.1+6
- Terminal-Bench HardClaude Opus 4.7 by 4.551.547+4.5
- SciCodeClaude Opus 4.7 by 2.454.552.1+2.4
- LiveBench · Agentic Codingtie50.750.3tie
AgentsClaude Opus 4.7
- τ-Bench V3 · BankingClaude Opus 4.7 by 23.534.611.1+23.5
- MCP AtlasClaude Opus 4.7 by 16.777.360.6+16.7
- BrowseCompClaude Opus 4.7 by 13.579.365.8+13.5
- AA ApexAgentsClaude Opus 4.7 by 10.933.923+10.9
- GDPValGPT-5.2 by 5.542.848.3+5.5
ReasoningClaude Opus 4.7
- ARC-AGI-2Claude Opus 4.7 by 22.975.852.9+22.9
- SimpleBenchClaude Opus 4.7 by 15.961.745.8+15.9
- Humanity's Last ExamClaude Opus 4.7 by 4.642.337.7+4.6
- LiveBench · ReasoningClaude Opus 4.7 by 487.283.2+4
- GPQA DiamondClaude Opus 4.7 by 1.191.490.3+1.1
- CritPttie1211.6tie
FactsClaude Opus 4.7
- AA-Omniscience · Non-hallucinationClaude Opus 4.7 by 38.957.718.8+38.9
- SimpleQA VerifiedClaude Opus 4.7 by 14.651.737.1+14.6
- AA-Omniscience · AccuracyClaude Opus 4.7 by 4.648.944.3+4.6
- Vectara HHEM hallucination ratelower is betterGPT-5.2 by 1.21210.8+1.2
Images and chartsClaude Opus 4.7
- CharXiv (RQ)Claude Opus 4.7 by 8.99182.1+8.9
- LMArena · VisionClaude Opus 4.7 by 67 rating points13131246+67 rating
- OCRBenchv2Claude Opus 4.7 by 6.456.950.5+6.4
- MMMU-ProClaude Opus 4.7 by 4.278.874.6+4.2
- MathVistaClaude Opus 4.7 by 1.684.482.8+1.6
MathGPT-5.2
- HMMT Feb. 2026GPT-5.2 by 3.193.997+3.1
- FrontierMath Tiers 1-3 (v2)Claude Opus 4.7 by 2.870.267.4+2.8
- AIME 2026GPT-5.2 by 2.595.898.3+2.5
- LiveBench · Mathematicstie92.893.2tie
- FrontierMath Tier 4tie31.731.7tie
Long documentsGPT-5.2
- MRCR v2 8 needle 128k (average)GPT-5.2 by 24.559.383.8+24.5
- AA-LCRGPT-5.2 by 478.782.7+4
Following instructionsEven
- IFBenchGPT-5.2 by 16.858.675.4+16.8
- LiveBench · Instruction FollowingClaude Opus 4.7 by 4.966.761.8+4.9
Other results21 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.
- GDPval-AA (Elo)Claude Opus 4.7 by 291 rating points17531462+291 rating
- AA-OmniscienceClaude Opus 4.7 by 28.227.3-0.9+28.2
- Artificial Analysis Coding IndexClaude Opus 4.7 by 24.973.648.7+24.9
- AA Agentic IndexGPT-5.2 by 20.739.560.2+20.7
- Terminal-Bench 2.0Claude Opus 4.7 by 15.469.454+15.4
- ToolathlonClaude Opus 4.7 by 1359.346.3+13
- AA IntelligenceClaude Opus 4.7 by 10.340.730.4+10.3
- HLE (with tools)Claude Opus 4.7 by 9.254.745.5+9.2
- WorldVQAClaude Opus 4.7 by 7.935.928+7.9
- ScreenSpot-Pro (No tools)GPT-5.2 by 6.879.586.3+6.8
- ARC-AGI-1Claude Opus 4.7 by 5.89286.2+5.8
- frontiermath_tier_4_v1Claude Opus 4.7 by 4.122.918.8+4.1
- τ²-Bench Telecom (AA run)Claude Opus 4.7 by 3.888.684.8+3.8
- vectara_avg_summary_lengthGPT-5.2 by 37.2 rating points149.1186.3+37.2 rating
- LiveBenchClaude Opus 4.7 by 2.176.974.8+2.1
- vectara_answer_rateGPT-5.2 by 298100+2
- livebench_languageGPT-5.2 by 1.977.979.8+1.9
- MMMLUClaude Opus 4.7 by 1.991.589.6+1.9
- vectara_factual_consistencyGPT-5.2 by 1.28889.2+1.2
- SimpleVQAtie56.555.8tie
- livebench_data_analysistie78.378.2tie
Questions people ask
Which is better, Claude Opus 4.7 or GPT-5.2?
Claude Opus 4.7 wins five of the eight areas we test: coding, agents, reasoning, facts and images and charts. GPT-5.2 wins math and long documents, and costs 48% less. They are level on following instructions.
Which is better for coding?
Claude Opus 4.7. It wins 7 of the 8 coding tests both models report; GPT-5.2 wins none, and 1 is a tie.
Which is cheaper?
Claude Opus 4.7 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 58 benchmark tests both models have published scores on. The verdict counts the 37 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 21 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.