Claude Sonnet 4.6 vs Kimi K2.6
Wins 3 of 8 areas
Coding · Agents · Math
Wins 5 of 8 areas
Reasoning · Facts · Images and charts · Long documents · Following instructions
Kimi K2.6 is the stronger all-rounder.Claude Sonnet 4.6 is better at agents.
Scores updated · 59 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.
- ReasoningHard problems that need careful thinking13Kimi K2.63 of 4 tests
- Images and chartsUnderstanding pictures, charts and video02Kimi K2.62 of 3 tests · 1 tie
- Following instructionsDoing exactly what it is asked02Kimi K2.62 of 2 tests
- FactsGetting facts right instead of making them up12Kimi K2.62 of 4 tests · 1 tie
- Long documentsFinding answers in very long texts01Kimi K2.61 of 1 test
- AgentsCarrying out multi-step tasks on its own42Claude Sonnet 4.64 of 8 tests · 2 ties
- MathCompetition and research-level math20Claude Sonnet 4.62 of 2 tests
- CodingWriting and fixing software32Claude Sonnet 4.63 of 8 tests · 3 ties
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.
Kimi K2.6 costs 73% 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 K2.6 pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+19.4points ahead
- Reasoning about charts from research papersCharXiv (RQ)+15.1points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+11.6points ahead
Where Claude Sonnet 4.6 pulls ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+11.1points ahead
- Real work tasks from 44 professionsGDPVal+9.7points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+9.1points ahead
Every test, side by side
All 59 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Sonnet 4.6
- Terminal-Bench HardClaude Sonnet 4.6 by 9.15343.9+9.1
- Terminal-Bench 2.1Claude Sonnet 4.6 by 5.371.265.9+5.3
- LiveBench · Agentic CodingKimi K2.6 by 4.342.646.9+4.3
- SciCodeKimi K2.6 by 1.450.151.5+1.4
- LMArena · WebDevClaude Sonnet 4.6 by 13 rating points15221509+13 rating
- LiveBench · Codingtie79.378.6tie
- SWE-bench Verifiedtie79.680.2tie
- SWE-bench Protie58.158.6tie
AgentsClaude Sonnet 4.6
- BrowseCompKimi K2.6 by 11.674.786.3+11.6
- τ-Bench V3 · BankingClaude Sonnet 4.6 by 11.134.423.3+11.1
- GDPValClaude Sonnet 4.6 by 9.736.727+9.7
- AA IT-Bench SREClaude Sonnet 4.6 by 8.639.831.2+8.6
- MCP AtlasKimi K2.6 by 6.861.368.1+6.8
- Terminal-Bench 4.0Claude Sonnet 4.6 by 2.530.5+2.5
- OSWorld-Verifiedtie72.573.1tie
- AA ApexAgentstie2828.5tie
ReasoningKimi K2.6
- LiveBench · ReasoningClaude Sonnet 4.6 by 5.484.879.4+5.4
- CritPtKimi K2.6 by 4.93.18+4.9
- Humanity's Last ExamKimi K2.6 by 3.933.637.5+3.9
- GPQA DiamondKimi K2.6 by 3.687.591.1+3.6
FactsKimi K2.6
- AA-Omniscience · AccuracyClaude Sonnet 4.6 by 8.340.932.6+8.3
- AA-Omniscience · Non-hallucinationKimi K2.6 by 7.951.659.5+7.9
- SimpleQA VerifiedKimi K2.6 by 2.132.834.9+2.1
- Vectara HHEM hallucination ratelower is bettertie10.610.8tie
Images and chartsKimi K2.6
- CharXiv (RQ)Kimi K2.6 by 15.171.686.7+15.1
- MMMU-ProKimi K2.6 by 6.173.379.4+6.1
- LMArena · Visiontie12831282tie
MathClaude Sonnet 4.6
- USAMO 2026Claude Sonnet 4.6 by 3.85551.2+3.8
- LiveBench · MathematicsClaude Sonnet 4.6 by 2.78784.3+2.7
Following instructionsKimi K2.6
- IFBenchKimi K2.6 by 19.456.676+19.4
- LiveBench · Instruction FollowingKimi K2.6 by 1.263.264.4+1.2
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.
- GDPval-AA v2Claude Sonnet 4.6 by 205 rating points13951190+205 rating
- τ²-Bench Telecom (AA run)Kimi K2.6 by 20.275.795.9+20.2
- GDPval-AA (Elo)Claude Sonnet 4.6 by 194 rating points16761482+194 rating
- livebench_data_analysisClaude Sonnet 4.6 by 12.97865.1+12.9
- AA Agentic IndexClaude Sonnet 4.6 by 1133.122.1+11
- ToolathlonKimi K2.6 by 94150+9
- Terminal-Bench 2.0Kimi K2.6 by 7.659.166.7+7.6
- HLE (with tools)Kimi K2.6 by 7.246.854+7.2
- AA-OmniscienceClaude Sonnet 4.6 by 6.912.25.3+6.9
- frontiermath_tier_4_v1Kimi K2.6 by 6.38.314.6+6.3
- Finance Agent v2Claude Sonnet 4.6 by 6.15144.9+6.1
- AgentWorldBench - SWEClaude Sonnet 4.6 by 5.764.558.8+5.7
- AgentWorldBench - MCPClaude Sonnet 4.6 by 4.87065.2+4.8
- AgentWorldBench - TerminalClaude Sonnet 4.6 by 4.55752.5+4.5
- LiveBenchClaude Sonnet 4.6 by 3.375.572.2+3.3
- AA IntelligenceClaude Sonnet 4.6 by 3.130.127+3.1
- AgentWorldBench - OverallClaude Sonnet 4.6 by 2.65653.4+2.6
- AgentWorldBench - OSClaude Sonnet 4.6 by 2.463.260.8+2.4
- AgentWorldBench - SearchClaude Sonnet 4.6 by 1.328.827.5+1.3
- Artificial Analysis Coding IndexClaude Sonnet 4.6 by 1.26361.8+1.2
- GPQA (unspecified)Kimi K2.6 by 1.189.991+1.1
- livebench_languageClaude Sonnet 4.6 by 176.175.1+1
- AgentWorldBench - Androidtie5858.9tie
- AgentWorldBench - Webtie50.850.2tie
- vectara_answer_ratetie99.999.7tie
- vectara_avg_summary_lengthtie114.7116.7tie
- vectara_factual_consistencytie89.489.2tie
Questions people ask
Which is better, Claude Sonnet 4.6 or Kimi K2.6?
Kimi K2.6 wins five of the eight areas we test: reasoning, facts, images and charts, long documents and following instructions. Claude Sonnet 4.6 wins coding, agents and math.
Which is better for coding?
Claude Sonnet 4.6. It wins 3 of the 8 coding tests both models report; Kimi K2.6 wins 2, and 3 are ties.
Which is cheaper?
Claude Sonnet 4.6 costs $3.00 per million input tokens and $15.00 per million output tokens; Kimi K2.6 costs $0.95 and $4.00. That makes Kimi K2.6 about 73% cheaper for the same work.
How do you compare the two?
We use the 59 benchmark tests both models have published scores on. The verdict counts the 32 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.