Claude Opus 4.8 vs Claude Opus 5
Wins 1 of 8 areas
Following instructions
Wins 6 of 8 areas
Coding · Agents · Reasoning · Facts · Math · Long documents
Claude Opus 5 is the stronger all-rounder.Claude Opus 4.8 is better at following instructions.
Both rank among the ten best models we track in coding and agents.
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 software07Claude Opus 57 of 8 tests · 1 tie
- ReasoningHard problems that need careful thinking07Claude Opus 57 of 7 tests
- AgentsCarrying out multi-step tasks on its own05Claude Opus 55 of 6 tests · 1 tie
- MathCompetition and research-level math03Claude Opus 53 of 3 tests
- FactsGetting facts right instead of making them up12Claude Opus 52 of 3 tests
- Long documentsFinding answers in very long texts01Claude Opus 51 of 1 test
- Following instructionsDoing exactly what it is asked10Claude Opus 4.81 of 1 test
- Images and chartsUnderstanding pictures, charts and video11Even1 each
Long documents and following instructions 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.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Claude Opus 5 pulls ahead
Where Claude Opus 4.8 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+21.5points ahead
- Follows detailed instructions exactlyLiveBench · Instruction Following+8.2points ahead
- Reasoning about charts from research papersCharXiv (RQ)+6.2points 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 5
- LiveBench · Agentic CodingClaude Opus 5 by 14.750.565.2+14.7
- LMArena · WebDevClaude Opus 5 by 136 rating points15561692+136 rating
- SWE-bench ProClaude Opus 5 by 1069.279.2+10
- SWE-bench VerifiedClaude Opus 5 by 7.488.696+7.4
- SWE-bench MultilingualClaude Opus 5 by 5.184.489.5+5.1
- Terminal-Bench 2.1Claude Opus 5 by 4.584.689.1+4.5
- SciCodeClaude Opus 5 by 254.456.4+2
- LiveBench · Codingtie81.881.5tie
AgentsClaude Opus 5
- Terminal-Bench 4.0Claude Opus 5 by 27.321.749+27.3
- GDPValClaude Opus 5 by 13.447.861.2+13.4
- τ-Bench V3 · BankingClaude Opus 5 by 7.934.242.1+7.9
- BrowseCompClaude Opus 5 by 6.584.390.8+6.5
- MCP AtlasClaude Opus 5 by 3.682.285.8+3.6
- OSWorld-Verifiedtie83.483.4tie
ReasoningClaude Opus 5
- ARC-AGI-3Claude Opus 5 by 28.71.530.2+28.7
- ARC-AGI-2Claude Opus 5 by 18.372.190.4+18.3
- SimpleBenchClaude Opus 5 by 15.864.880.6+15.8
- CritPtClaude Opus 5 by 8.220.929.1+8.2
- Humanity's Last ExamClaude Opus 5 by 6.248.754.9+6.2
- LiveBench · ReasoningClaude Opus 5 by 289.291.2+2
- GPQA DiamondClaude Opus 5 by 1.29293.2+1.2
FactsClaude Opus 5
- AA-Omniscience · Non-hallucinationClaude Opus 4.8 by 21.560.739.2+21.5
- AA-Omniscience · AccuracyClaude Opus 5 by 12.148.860.9+12.1
- SimpleQA VerifiedClaude Opus 5 by 6.95359.9+6.9
Images and chartsEven
- CharXiv (RQ)Claude Opus 4.8 by 6.289.983.7+6.2
- LMArena · VisionClaude Opus 5 by 26 rating points12951321+26 rating
MathClaude Opus 5
- FrontierMath Tier 4Claude Opus 5 by 17.156.173.2+17.1
- FrontierMath Tiers 1-3 (v2)Claude Opus 5 by 5.68085.6+5.6
- LiveBench · MathematicsClaude Opus 5 by 1.494.395.7+1.4
Following instructionsClaude Opus 4.8
- LiveBench · Instruction FollowingClaude Opus 4.8 by 8.27263.8+8.2
Other results27 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.
- Terminal-Bench 4.0Claude Opus 5 by 28.223.651.8+28.2
- GDPval-AA v2Claude Opus 5 by 231 rating points15931824+231 rating
- GDPval-AA v2 EloClaude Opus 5 by 231 rating points15931824+231 rating
- HiL-Bench (Tools-allowed)Claude Opus 5 by 21.735.357+21.7
- SWE-bench MultimodalClaude Opus 5 by 2138.459.4+21
- GDPval-AA (Elo)Claude Opus 4.8 by 182 rating points18901708+182 rating
- JobBenchClaude Opus 5 by 17.348.465.7+17.3
- OSWorld 2.0Claude Opus 5 by 14.955.770.6+14.9
- AA Agentic IndexClaude Opus 5 by 13.642.656.2+13.6
- Program BenchClaude Opus 5 by 13.571.985.4+13.5
- HealthBench Professional (raw)Claude Opus 5 by 13.160.373.4+13.1
- DeepSWE 1.1Claude Opus 5 by 9.85968.8+9.8
- AA IntelligenceClaude Opus 5 by 941.850.8+9
- livebench_languageClaude Opus 5 by 979.788.7+9
- livebench_data_analysisClaude Opus 5 by 8.66674.5+8.6
- AA-OmniscienceClaude Opus 5 by 8.328.837.1+8.3
- HealthBench (raw)Claude Opus 5 by 8.358.867.1+8.3
- Agents' Last ExamClaude Opus 5 by 5.925.731.6+5.9
- HLE (with tools)Claude Opus 5 by 5.757.963.6+5.7
- ARC-AGI-1Claude Opus 5 by 592.597.5+5
- HealthBench ProfessionalClaude Opus 5 by 455.859.8+4
- Artificial Analysis Coding IndexClaude Opus 5 by 3.774.378+3.7
- Legal Agent BenchmarkClaude Opus 5 by 1.71011.7+1.7
- GDP (Surge AI)tie84.885.5tie
- OfficeQA Protie66.266.9tie
- Toolathlon Verifiedtie79.980.6tie
- OfficeQAtie77.678.1tie
Questions people ask
Which is better, Claude Opus 4.8 or Claude Opus 5?
Claude Opus 5 wins six of the eight areas we test: coding, agents, reasoning, facts, math and long documents. Claude Opus 4.8 wins following instructions. They are level on images and charts.
Which is better for coding?
Claude Opus 5. It wins 7 of the 8 coding tests both models report; Claude Opus 4.8 wins none, and 1 is a tie.
Which is cheaper?
Claude Opus 4.8 costs $5.00 per million input tokens and $25.00 per million output tokens; Claude Opus 5 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 58 benchmark tests both models have published scores on. The verdict counts the 31 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 27 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.