Claude Opus 4.8 vs Kimi K3
Wins 2 of 8 areas
Facts · Math
Wins 4 of 8 areas
Coding · Agents · Images and charts · Long documents
Kimi K3 wins more areas, narrowly.Claude Opus 4.8 is better at math.
Both rank among the ten best models we track in agents.
Scores updated · 64 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.
- AgentsCarrying out multi-step tasks on its own16Kimi K36 of 7 tests
- CodingWriting and fixing software03Kimi K33 of 5 tests · 2 ties
- Images and chartsUnderstanding pictures, charts and video01Kimi K31 of 1 test
- Long documentsFinding answers in very long texts01Kimi K31 of 1 test
- MathCompetition and research-level math41Claude Opus 4.84 of 5 tests
- FactsGetting facts right instead of making them up30Claude Opus 4.83 of 3 tests
- ReasoningHard problems that need careful thinking33Even3 each
- Following instructionsDoing exactly what it is asked00Even0 each · 1 tie
Images and charts, 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.
Kimi K3 costs 40% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Kimi K3 pulls ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+11.8points ahead
- Fixes real GitHub issues, working as an agentLiveBench · Agentic Coding+11.7points ahead
- Reasons across sets of long documentsAA-LCR+11points ahead
Where Claude Opus 4.8 pulls ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+17.1points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+13.9points ahead
- Abstract visual puzzles that people can solveARC-AGI-2+11.7points ahead
Every test, side by side
All 64 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingKimi K3
- LiveBench · Agentic CodingKimi K3 by 11.750.562.2+11.7
- LMArena · WebDevKimi K3 by 99 rating points15561655+99 rating
- SciCodeKimi K3 by 5.154.459.5+5.1
- Terminal-Bench 2.1tie84.685tie
- LiveBench · Codingtie81.881.5tie
AgentsKimi K3
- τ-Bench V3 · BankingKimi K3 by 11.834.246+11.8
- Terminal-Bench 4.0Claude Opus 4.8 by 9.121.712.6+9.1
- BrowseCompKimi K3 by 6.984.391.2+6.9
- GDPValKimi K3 by 3.947.851.7+3.9
- MCP AtlasKimi K3 by 282.284.2+2
- AA ApexAgentsKimi K3 by 1.939.441.3+1.9
- OSWorld-VerifiedKimi K3 by 1.483.484.8+1.4
ReasoningEven
- ARC-AGI-2Claude Opus 4.8 by 11.772.160.4+11.7
- SimpleBenchClaude Opus 4.8 by 4.164.860.7+4.1
- CritPtKimi K3 by 2.520.923.4+2.5
- Humanity's Last ExamClaude Opus 4.8 by 1.848.746.9+1.8
- GPQA DiamondKimi K3 by 1.59293.5+1.5
- LiveBench · ReasoningKimi K3 by 1.589.290.7+1.5
FactsClaude Opus 4.8
- AA-Omniscience · Non-hallucinationClaude Opus 4.8 by 13.960.746.8+13.9
- SimpleQA VerifiedClaude Opus 4.8 by 2.45350.6+2.4
- AA-Omniscience · AccuracyClaude Opus 4.8 by 1.248.847.6+1.2
MathClaude Opus 4.8
- FrontierMath Tier 4Claude Opus 4.8 by 17.156.139+17.1
- LiveBench · MathematicsClaude Opus 4.8 by 9.994.384.4+9.9
- FrontierMath Tiers 1-3 (v2)Claude Opus 4.8 by 7.88072.2+7.8
- AIME 2026Claude Opus 4.8 by 3.310096.7+3.3
- HMMT Feb. 2026Kimi K3 by 1.595.597+1.5
Following instructionsEven
- LiveBench · Instruction Followingtie7271.4tie
Other results35 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.
- MCP Mark VerifiedKimi K3 by 18.176.494.5+18.1
- SWE-MarathonKimi K3 by 162642+16
- ToolathlonKimi K3 by 13.359.973.2+13.3
- livebench_data_analysisKimi K3 by 12.76678.7+12.7
- NL2RepoClaude Opus 4.8 by 11.769.758+11.7
- DeepSWE 1.1Kimi K3 by 105969+10
- matharena_visual_math_overallKimi K3 by 9.781.691.3+9.7
- DeepSWEKimi K3 by 9.55867.5+9.5
- GDPval-AA v2 EloKimi K3 by 93 rating points15931686+93 rating
- AA-OmniscienceClaude Opus 4.8 by 9.128.819.7+9.1
- GDPval-AA v2Kimi K3 by 89 rating points15931682+89 rating
- DSBench-HardClaude Opus 4.8 by 8.771.763+8.7
- AA Agentic IndexKimi K3 by 842.650.6+8
- Program BenchKimi K3 by 5.971.977.8+5.9
- livebench_languageKimi K3 by 5.879.785.5+5.8
- MLS Bench Litelower is betterClaude Opus 4.8 by 5.542.848.3+5.5
- JobBenchKimi K3 by 4.548.452.9+4.5
- Terminal-bench 3.0Claude Opus 4.8 by 3.721.117.4+3.7
- AutomationBench PublicKimi K3 by 3.627.230.8+3.6
- Toolathlon VerifiedClaude Opus 4.8 by 3.479.976.5+3.4
- SpreadsheetBench 2Kimi K3 by 3.231.634.8+3.2
- OfficeQA ProClaude Opus 4.8 by 2.966.263.3+2.9
- ResearchRubricsKimi K3 by 2.773.576.2+2.7
- Agents' Last ExamKimi K3 by 2.625.728.3+2.6
- DSBench-FullStackKimi K3 by 2.171.673.7+2.1
- ARC-AGI-1Kimi K3 by 292.594.5+2
- ProgramBench (Almost Solved)Kimi K3 by 215.517.5+2
- Agents' Last Exam (ALE-CLI)Kimi K3 by 1.925.727.6+1.9
- Artificial Analysis Coding IndexKimi K3 by 1.974.376.2+1.9
- DeepSearchQA (F1)Kimi K3 by 1.993.195+1.9
- HLE (with tools)Kimi K3 by 1.957.959.8+1.9
- HLE (wo / w tools)Claude Opus 4.8 by 1.957.956+1.9
- AA IntelligenceKimi K3 by 1.841.843.6+1.8
- CyberGymKimi K3 by 1.278.880+1.2
- PostTrainBenchtie37.236.6tie
Questions people ask
Which is better, Claude Opus 4.8 or Kimi K3?
Kimi K3 wins four of the eight areas we test: coding, agents, images and charts and long documents. Claude Opus 4.8 wins facts and math. They are level on reasoning and following instructions.
Which is better for coding?
Kimi K3. It wins 3 of the 5 coding tests both models report; Claude Opus 4.8 wins none, 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; Kimi K3 costs $3.00 and $15.00. That makes Kimi K3 about 40% cheaper for the same work.
How do you compare the two?
We use the 64 benchmark tests both models have published scores on. The verdict counts the 29 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 35 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.