Claude Sonnet 4 vs Kimi K2 Instruct
Wins 5 of 6 areas
Coding · Reasoning · Facts · Long documents · Following instructions
Wins 1 of 6 areas
Agents
Claude Sonnet 4 is the stronger all-rounder.Kimi K2 Instruct is cheaper and better at agents.
Scores updated · 63 tests both models report · How we compare
Where each one wins
Tests won in each of the six areas where both have results. 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 software51Claude Sonnet 45 of 6 tests
- ReasoningHard problems that need careful thinking20Claude Sonnet 42 of 3 tests · 1 tie
- Following instructionsDoing exactly what it is asked20Claude Sonnet 42 of 2 tests
- FactsGetting facts right instead of making them up21Claude Sonnet 42 of 3 tests
- Long documentsFinding answers in very long texts10Claude Sonnet 41 of 1 test
- AgentsCarrying out multi-step tasks on its own02Kimi K2 Instruct2 of 2 tests
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 Instruct costs 83% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Claude Sonnet 4 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+40.6points ahead
- Reasons across sets of long documentsAA-LCR+17.3points ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+15points ahead
Where Kimi K2 Instruct pulls ahead
- Real work tasks from 44 professionsGDPVal+8.6points ahead
- Recent programming contest problemsLiveCodeBench v6+5.2points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+2.7points ahead
Every test, side by side
All 63 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Sonnet 4
- SWE-bench ProClaude Sonnet 4 by 1542.727.7+15
- SWE-bench MultilingualClaude Sonnet 4 by 9.656.947.3+9.6
- SciCodeClaude Sonnet 4 by 9.34030.7+9.3
- Terminal-Bench HardClaude Sonnet 4 by 7.631.123.5+7.6
- SWE-bench VerifiedClaude Sonnet 4 by 6.972.765.8+6.9
- LiveCodeBench v6Kimi K2 Instruct by 5.248.553.7+5.2
AgentsKimi K2 Instruct
- GDPValKimi K2 Instruct by 8.69.618.2+8.6
- BrowseCompKimi K2 Instruct by 1.912.214.1+1.9
ReasoningClaude Sonnet 4
- Humanity's Last ExamClaude Sonnet 4 by 4.310.76.4+4.3
- GPQA DiamondClaude Sonnet 4 by 177.776.7+1
- CritPttie0.30tie
FactsClaude Sonnet 4
- AA-Omniscience · Non-hallucinationClaude Sonnet 4 by 40.670.930.3+40.6
- Vectara HHEM hallucination ratelower is betterClaude Sonnet 4 by 7.610.317.9+7.6
- AA-Omniscience · AccuracyKimi K2 Instruct by 2.722.725.4+2.7
Following instructionsClaude Sonnet 4
- IFBenchClaude Sonnet 4 by 1354.741.7+13
- Multi-ChallengeClaude Sonnet 4 by 357.154.1+3
Other results46 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.
- AA-OmniscienceClaude Sonnet 4 by 26.80.2-26.6+26.8
- AIME 2024Kimi K2 Instruct by 26.243.469.6+26.2
- HMMT 2025Kimi K2 Instruct by 22.915.938.8+22.9
- AIME 2025Claude Sonnet 4 by 2170.549.5+21
- τ²-BenchKimi K2 Instruct by 20.645.265.8+20.6
- AA Agentic IndexKimi K2 Instruct by 20.117.637.7+20.1
- ZebraLogicKimi K2 Instruct by 15.373.789+15.3
- SimpleQAKimi K2 Instruct by 15.115.931+15.1
- LiveCodeBenchClaude Sonnet 4 by 14.868.553.7+14.8
- AIR-Bench 2024Claude Sonnet 4 by 14.288.374.1+14.2
- CNMO 2024Kimi K2 Instruct by 13.960.474.3+13.9
- FinSearchComp-globalClaude Sonnet 4 by 12.54229.5+12.5
- PolyMath-enKimi K2 Instruct by 12.352.865.1+12.3
- OJBenchKimi K2 Instruct by 11.815.327.1+11.8
- Artificial Analysis Coding IndexClaude Sonnet 4 by 11.737.625.9+11.7
- Aider-PolyglotClaude Sonnet 4 by 10.770.760+10.7
- τ²-Bench Telecom (AA run)Kimi K2 Instruct by 8.864.673.4+8.8
- vectara_avg_summary_lengthClaude Sonnet 4 by 86.6 rating points145.859.2+86.6 rating
- GAIA (text only)Claude Sonnet 4 by 8.168.360.2+8.1
- vectara_factual_consistencyClaude Sonnet 4 by 7.689.782.1+7.6
- theagentcompanyClaude Sonnet 4 by 73730+7
- HLE (with tools)Kimi K2 Instruct by 6.620.326.9+6.6
- Terminal-BenchClaude Sonnet 4 by 5.535.530+5.5
- SWE-Dev (Tools-allowed)Claude Sonnet 4 by 5.267.161.9+5.2
- τ²-Bench (Retail)Claude Sonnet 4 by 4.47570.6+4.4
- AA IntelligenceClaude Sonnet 4 by 3.618.915.3+3.6
- xbench-DeepSearchClaude Sonnet 4 by 3.664.661+3.6
- MATH-500 (EM)Kimi K2 Instruct by 3.49497.4+3.4
- ArtifactsBenchClaude Sonnet 4 by 3.157.354.2+3.1
- MultiPL-EClaude Sonnet 4 by 2.988.685.7+2.9
- MMLU-ProClaude Sonnet 4 by 2.683.781.1+2.6
- IFEvalKimi K2 Instruct by 2.287.689.8+2.2
- Multi-SWE-BenchClaude Sonnet 4 by 2.235.733.5+2.2
- MMLUClaude Sonnet 4 by 291.589.5+2
- LiveBenchKimi K2 Instruct by 1.674.876.4+1.6
- XSTestKimi K2 Instruct by 1.696.698.2+1.6
- SuperGPQAKimi K2 Instruct by 1.555.757.2+1.5
- Tau2 airlineKimi K2 Instruct by 155.556.5+1
- MMLU-Reduxtie93.692.7tie
- HarmBenchtie9897.4tie
- AceBenchtie76.276.5tie
- AutoLogitie89.889.5tie
- BrowseComp-ZHtie29.128.8tie
- anthropic_red_teamtie99.299.3tie
- simple_safety_teststie100100tie
- vectara_answer_ratetie98.698.6tie
Questions people ask
Which is better, Claude Sonnet 4 or Kimi K2 Instruct?
Claude Sonnet 4 wins five of the six areas where both have results: coding, reasoning, facts, long documents and following instructions. Kimi K2 Instruct wins agents, and costs 83% less.
Which is better for coding?
Claude Sonnet 4. It wins 5 of the 6 coding tests both models report; Kimi K2 Instruct wins 1.
Which is cheaper?
Claude Sonnet 4 costs $3.00 per million input tokens and $15.00 per million output tokens; Kimi K2 Instruct costs $0.60 and $2.50. That makes Kimi K2 Instruct about 83% cheaper for the same work.
How do you compare the two?
We use the 63 benchmark tests both models have published scores on. The verdict counts the 17 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 46 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.