Claude Opus 4.8 vs GPT-5.5
Wins 1 of 8 areas
Agents
Wins 5 of 8 areas
Coding · Reasoning · Facts · Math · Long documents
GPT-5.5 is the stronger all-rounder.Claude Opus 4.8 is cheaper and better at agents.
Both rank among the ten best models we track in coding and agents.
Scores updated · 83 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 math05GPT-5.55 of 6 tests · 1 tie
- ReasoningHard problems that need careful thinking24GPT-5.54 of 7 tests · 1 tie
- CodingWriting and fixing software23GPT-5.53 of 7 tests · 2 ties
- FactsGetting facts right instead of making them up12GPT-5.52 of 3 tests
- Long documentsFinding answers in very long texts01GPT-5.51 of 1 test
- AgentsCarrying out multi-step tasks on its own51Claude Opus 4.85 of 7 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video00Even0 each · 1 tie
- Following instructionsDoing exactly what it is asked11Even1 each
Images and charts and long documents rest on a single test 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.
Claude Opus 4.8 costs 14% 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-5.5 pulls ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+16.4points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+13.7points ahead
- Abstract visual puzzles that people can solveARC-AGI-2+12.9points ahead
Where Claude Opus 4.8 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+49.7points ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+10.6points ahead
- Complex command-line tasks across many fieldsTerminal-Bench 4.0+7.1points ahead
Every test, side by side
All 83 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5.5
- SWE-bench ProClaude Opus 4.8 by 10.669.258.6+10.6
- LMArena · WebDevClaude Opus 4.8 by 43 rating points15561513+43 rating
- LiveBench · Agentic CodingGPT-5.5 by 3.550.554+3.5
- Terminal-Bench HardGPT-5.5 by 2.358.360.6+2.3
- SciCodeGPT-5.5 by 1.454.455.8+1.4
- LiveBench · Codingtie81.882.2tie
- Terminal-Bench 2.1tie84.684.3tie
AgentsClaude Opus 4.8
- Terminal-Bench 4.0Claude Opus 4.8 by 7.121.714.6+7.1
- MCP AtlasClaude Opus 4.8 by 6.982.275.3+6.9
- GDPValClaude Opus 4.8 by 5.147.842.7+5.1
- τ-Bench V3 · BankingGPT-5.5 by 4.834.239+4.8
- OSWorld-VerifiedClaude Opus 4.8 by 4.783.478.7+4.7
- AA ApexAgentsClaude Opus 4.8 by 1.739.437.7+1.7
- BrowseComptie84.384.4tie
ReasoningGPT-5.5
- ARC-AGI-2GPT-5.5 by 12.972.185+12.9
- CritPtGPT-5.5 by 6.220.927.1+6.2
- SimpleBenchGPT-5.5 by 4.264.869+4.2
- Humanity's Last ExamClaude Opus 4.8 by 2.948.745.8+2.9
- GPQA DiamondGPT-5.5 by 1.59293.5+1.5
- ARC-AGI-3Claude Opus 4.8 by 1.11.50.4+1.1
- LiveBench · Reasoningtie89.289.7tie
FactsGPT-5.5
- AA-Omniscience · Non-hallucinationClaude Opus 4.8 by 49.760.711+49.7
- SimpleQA VerifiedGPT-5.5 by 105363+10
- AA-Omniscience · AccuracyGPT-5.5 by 9.248.858+9.2
MathGPT-5.5
- FrontierMath Tier 4GPT-5.5 by 16.456.172.5+16.4
- FrontierMath Tiers 1-3 (v2)GPT-5.5 by 5.38085.3+5.3
- HMMT Feb. 2026GPT-5.5 by 395.598.5+3
- LiveBench · MathematicsGPT-5.5 by 1.694.395.9+1.6
- USAMO 2026GPT-5.5 by 1.596.798.2+1.5
- AIME 2026tie100100tie
Following instructionsEven
- IFBenchGPT-5.5 by 13.762.275.9+13.7
- LiveBench · Instruction FollowingClaude Opus 4.8 by 1.37270.7+1.3
Other results49 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.
- GDP (Surge AI)Claude Opus 4.8 by 59.984.824.9+59.9
- GraphWalks Parents 1M subsetClaude Opus 4.8 by 24.883.358.5+24.8
- GraphWalks BFS 1M subsetClaude Opus 4.8 by 22.768.145.4+22.7
- NL2RepoClaude Opus 4.8 by 1969.750.7+19
- HLE CalibrationGPT-5.5 by 17.726.544.2+17.7
- MCP Mark VerifiedGPT-5.5 by 16.576.492.9+16.5
- livebench_data_analysisGPT-5.5 by 15.66681.6+15.6
- SWE-MarathonClaude Opus 4.8 by 142612+14
- matharena_visual_math_overallGPT-5.5 by 13.381.694.9+13.3
- GraphWalks BFS 256KClaude Opus 4.8 by 12.285.973.7+12.2
- GDPval-AA (Elo)Claude Opus 4.8 by 121 rating points18901769+121 rating
- OfficeQA ProClaude Opus 4.8 by 12.166.254.1+12.1
- DeepSWEGPT-5.5 by 125870+12
- GDPval-AA v2 EloClaude Opus 4.8 by 102 rating points15931491+102 rating
- JobBenchClaude Opus 4.8 by 10.148.438.3+10.1
- ResearchRubricsClaude Opus 4.8 by 9.573.564+9.5
- GraphWalks — Parents task, 256K-token context subsetClaude Opus 4.8 by 9.299.390.1+9.2
- PostTrainBenchClaude Opus 4.8 by 8.837.228.4+8.8
- GDPval-AA v2Claude Opus 4.8 by 84 rating points15931509+84 rating
- AA-OmniscienceClaude Opus 4.8 by 8.328.820.5+8.3
- Terminal-Bench 2.0GPT-5.5 by 8.174.682.7+8.1
- DeepSWE 1.1GPT-5.5 by 85967+8
- Legal Agent BenchmarkClaude Opus 4.8 by 7.9102.1+7.9
- livebench_languageGPT-5.5 by 7.779.787.4+7.7
- MLS Bench Litelower is betterGPT-5.5 by 7.342.835.5+7.3
- Toolathlon VerifiedClaude Opus 4.8 by 6.479.973.5+6.4
- Finance AgentGPT-5.5 by 6.153.960+6.1
- HLE (with tools)Claude Opus 4.8 by 5.757.952.2+5.7
- AA Agentic IndexClaude Opus 4.8 by 5.342.637.3+5.3
- HiL-Bench (Tools-allowed)GPT-5.5 by 4.435.339.7+4.4
- Tool-DecathlonClaude Opus 4.8 by 4.359.955.6+4.3
- ToolathlonClaude Opus 4.8 by 4.359.955.6+4.3
- frontiermath_tier_4_v1GPT-5.5 by 4.131.335.4+4.1
- HealthBench ProfessionalClaude Opus 4.8 by 455.851.8+4
- LiveBenchGPT-5.5 by 3.577.280.7+3.5
- AA IntelligenceClaude Opus 4.8 by 3.441.838.4+3.4
- OR-Bench (FRR)GPT-5.5 by 3.43.36.7+3.4
- CyberGymGPT-5.5 by 378.881.8+3
- HealthBenchClaude Opus 4.8 by 2.859.356.5+2.8
- ARC-AGI-1GPT-5.5 by 2.592.595+2.5
- SpreadsheetBench 2Claude Opus 4.8 by 2.531.629.1+2.5
- Finance Agent v2Claude Opus 4.8 by 2.153.951.8+2.1
- AIRS-BenchGPT-5.5 by 28486+2
- Program BenchClaude Opus 4.8 by 1.171.970.8+1.1
- Agents' Last Examtie25.726.6tie
- Artificial Analysis Coding Indextie74.374.9tie
- τ²-Bench Telecom (AA run)tie94.493.9tie
- BrowseComp (single-agent)tie84.384.4tie
- HMMT Nov. 2025tie96.596.5tie
Questions people ask
Which is better, Claude Opus 4.8 or GPT-5.5?
GPT-5.5 wins five of the eight areas we test: coding, reasoning, facts, math and long documents. Claude Opus 4.8 wins agents, and costs 14% less. They are level on images and charts and following instructions.
Which is better for coding?
GPT-5.5. It wins 3 of the 7 coding tests both models report; Claude Opus 4.8 wins 2, and 2 are ties.
Which is cheaper?
Claude Opus 4.8 costs $5.00 per million input tokens and $25.00 per million output tokens; GPT-5.5 costs $5.00 and $30.00. That makes Claude Opus 4.8 about 14% cheaper for the same work.
How do you compare the two?
We use the 83 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 49 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.