Claude Opus 4.6 vs Claude Opus 4.8
Wins 0 of 8 areas
—
Wins 6 of 8 areas
Coding · Agents · Reasoning · Facts · Math · Following instructions
Claude Opus 4.8 is the stronger all-rounder.
Scores updated · 77 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 software08Claude Opus 4.88 of 8 tests
- MathCompetition and research-level math06Claude Opus 4.86 of 7 tests · 1 tie
- ReasoningHard problems that need careful thinking15Claude Opus 4.85 of 7 tests · 1 tie
- FactsGetting facts right instead of making them up03Claude Opus 4.83 of 3 tests
- AgentsCarrying out multi-step tasks on its own13Claude Opus 4.83 of 5 tests · 1 tie
- Following instructionsDoing exactly what it is asked02Claude Opus 4.82 of 2 tests
- Images and chartsUnderstanding pictures, charts and video11Even1 each
- Long documentsFinding answers in very long texts00Even0 each · 1 tie
Long documents 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.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Claude Opus 4.8 pulls ahead
- Proof problems from the US Math OlympiadUSAMO 2026+30.5points ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+29.3points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+23.5points ahead
Where Claude Opus 4.6 pulls ahead
- Real work tasks from 44 professionsGDPVal+8.1points ahead
- Common-sense trick questionsSimpleBench+2.8points ahead
Every test, side by side
All 77 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Opus 4.8
- SWE-bench ProClaude Opus 4.8 by 15.853.469.2+15.8
- Terminal-Bench HardClaude Opus 4.8 by 12.146.258.3+12.1
- SWE-bench VerifiedClaude Opus 4.8 by 7.880.888.6+7.8
- SWE-bench MultilingualClaude Opus 4.8 by 6.677.884.4+6.6
- LiveBench · CodingClaude Opus 4.8 by 3.678.281.8+3.6
- SciCodeClaude Opus 4.8 by 2.551.954.4+2.5
- LiveBench · Agentic CodingClaude Opus 4.8 by 1.54950.5+1.5
- LMArena · WebDevClaude Opus 4.8 by 10 rating points15461556+10 rating
AgentsClaude Opus 4.8
- MCP AtlasClaude Opus 4.8 by 19.562.782.2+19.5
- OSWorld-VerifiedClaude Opus 4.8 by 10.772.783.4+10.7
- GDPValClaude Opus 4.6 by 8.155.947.8+8.1
- AA ApexAgentsClaude Opus 4.8 by 6.43339.4+6.4
- BrowseComptie8484.3tie
ReasoningClaude Opus 4.8
- Humanity's Last ExamClaude Opus 4.8 by 8.839.948.7+8.8
- CritPtClaude Opus 4.8 by 8.312.620.9+8.3
- ARC-AGI-2Claude Opus 4.8 by 3.368.872.1+3.3
- SimpleBenchClaude Opus 4.6 by 2.867.664.8+2.8
- GPQA DiamondClaude Opus 4.8 by 2.489.692+2.4
- ARC-AGI-3Claude Opus 4.8 by 10.51.5+1
- LiveBench · Reasoningtie88.789.2tie
FactsClaude Opus 4.8
- AA-Omniscience · Non-hallucinationClaude Opus 4.8 by 23.537.260.7+23.5
- SimpleQA VerifiedClaude Opus 4.8 by 64753+6
- AA-Omniscience · AccuracyClaude Opus 4.8 by 1.84748.8+1.8
Images and chartsEven
- CharXiv (RQ)Claude Opus 4.8 by 12.577.489.9+12.5
- LMArena · VisionClaude Opus 4.6 by 21 rating points13161295+21 rating
MathClaude Opus 4.8
- USAMO 2026Claude Opus 4.8 by 30.566.296.7+30.5
- FrontierMath Tier 4Claude Opus 4.8 by 29.326.856.1+29.3
- FrontierMath Tiers 1-3 (v2)Claude Opus 4.8 by 146680+14
- IMOAnswerBenchClaude Opus 4.8 by 8.275.383.5+8.2
- LiveBench · MathematicsClaude Opus 4.8 by 589.394.3+5
- AIME 2026Claude Opus 4.8 by 3.396.7100+3.3
- HMMT Feb. 2026tie96.295.5tie
Following instructionsClaude Opus 4.8
- IFBenchClaude Opus 4.8 by 9.153.162.2+9.1
- LiveBench · Instruction FollowingClaude Opus 4.8 by 8.763.372+8.7
Other results42 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.
- ScreenSpot-Pro (No tools)Claude Opus 4.8 by 30.257.787.9+30.2
- GDPval-AA (Elo)Claude Opus 4.8 by 271 rating points16191890+271 rating
- Artificial Analysis Coding IndexClaude Opus 4.8 by 26.248.174.3+26.2
- AA Agentic IndexClaude Opus 4.6 by 2567.642.6+25
- NL2RepoClaude Opus 4.8 by 19.949.869.7+19.9
- AA-OmniscienceClaude Opus 4.8 by 15.113.728.8+15.1
- AgentWorldBench - MCPClaude Opus 4.6 by 1569.954.9+15
- Tool-DecathlonClaude Opus 4.8 by 12.747.259.9+12.7
- JobBenchClaude Opus 4.8 by 11.836.648.4+11.8
- SWE-bench MultimodalClaude Opus 4.8 by 11.327.138.4+11.3
- AA IntelligenceClaude Opus 4.8 by 9.931.941.8+9.9
- matharena_visual_math_overallClaude Opus 4.8 by 9.372.381.6+9.3
- Terminal-Bench 2.0Claude Opus 4.8 by 9.265.474.6+9.2
- OfficeQA ProClaude Opus 4.8 by 9.157.166.2+9.1
- frontiermath_tier_4_v1Claude Opus 4.8 by 8.422.931.3+8.4
- Toolathlon Avg turnsClaude Opus 4.8 by 7.616.924.5+7.6
- Finance AgentClaude Opus 4.6 by 6.860.753.9+6.8
- AgentWorldBench - SearchClaude Opus 4.8 by 5.829.335.1+5.8
- Legal Agent BenchmarkClaude Opus 4.8 by 5.84.210+5.8
- CharXiv Reasoning (With tools)Claude Opus 4.8 by 5.284.789.9+5.2
- CyberGymClaude Opus 4.8 by 573.878.8+5
- ScreenSpot-Pro (With tools)Claude Opus 4.8 by 4.883.187.9+4.8
- HLE (with tools)Claude Opus 4.8 by 4.653.357.9+4.6
- OfficeQAClaude Opus 4.8 by 4.173.577.6+4.1
- livebench_data_analysisClaude Opus 4.6 by 3.969.966+3.9
- AgentWorldBench - OSClaude Opus 4.6 by 3.670.266.6+3.6
- livebench_languageClaude Opus 4.6 by 3.683.379.7+3.6
- AgentWorldBench - WebClaude Opus 4.8 by 3.351.454.7+3.3
- ToolathlonClaude Opus 4.8 by 3.156.859.9+3.1
- HiL-Bench (Tools-allowed)Claude Opus 4.6 by 338.335.3+3
- τ²-Bench Telecom (AA run)Claude Opus 4.8 by 2.392.194.4+2.3
- BrowseComp (multi-agent)Claude Opus 4.8 by 1.986.688.5+1.9
- DeepSearchQA (F1)Claude Opus 4.8 by 1.891.393.1+1.8
- AgentWorldBench - TerminalClaude Opus 4.8 by 1.757.559.2+1.7
- AgentWorldBench - OverallClaude Opus 4.6 by 1.257.856.6+1.2
- LiveBenchtie76.377.2tie
- Toolathlon Pass@∞tie47.248.1tie
- BrowseComp (single-agent)tie83.784.3tie
- AgentWorldBench - SWEtie64.564.1tie
- ARC-AGI-1tie9392.5tie
- AgentWorldBench - Androidtie61.761.5tie
- HMMT Nov. 2025tie96.396.5tie
Questions people ask
Which is better, Claude Opus 4.6 or Claude Opus 4.8?
Claude Opus 4.8 wins six of the eight areas we test: coding, agents, reasoning, facts, math and following instructions. Claude Opus 4.6 wins none. They are level on images and charts and long documents.
Which is better for coding?
Claude Opus 4.8. It wins 8 of the 8 coding tests both models report; Claude Opus 4.6 wins none.
Which is cheaper?
Claude Opus 4.6 costs $5.00 per million input tokens and $25.00 per million output tokens; Claude Opus 4.8 costs $5.00 and $25.00. For a million tokens read plus a million written, they cost about the same.
How do you compare the two?
We use the 77 benchmark tests both models have published scores on. The verdict counts the 35 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 42 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.