Kimi K2.5 vs Kimi K2.6
Wins 0 of 8 areas
—
Wins 8 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Math · Long documents · Following instructions
Kimi K2.6 is the stronger all-rounder.Kimi K2.5 is cheaper.
Scores updated · 63 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 software08Kimi K2.68 of 8 tests
- AgentsCarrying out multi-step tasks on its own06Kimi K2.66 of 6 tests
- ReasoningHard problems that need careful thinking03Kimi K2.63 of 3 tests
- Images and chartsUnderstanding pictures, charts and video03Kimi K2.63 of 3 tests
- MathCompetition and research-level math02Kimi K2.62 of 3 tests · 1 tie
- 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
- Following instructionsDoing exactly what it is asked01Kimi K2.61 of 1 test
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 K2.5 costs 27% less for the same work.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Kimi K2.6 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+25.2points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+20.2points ahead
- Long professional tasks in banking, consulting and lawAA ApexAgents+17points ahead
Where Kimi K2.5 pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+2.6points 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.
CodingKimi K2.6
- Terminal-Bench 2.1Kimi K2.6 by 20.245.765.9+20.2
- Terminal-Bench HardKimi K2.6 by 9.134.843.9+9.1
- SWE-bench ProKimi K2.6 by 7.950.758.6+7.9
- LMArena · WebDevKimi K2.6 by 73 rating points14361509+73 rating
- LiveCodeBench v6Kimi K2.6 by 4.68589.6+4.6
- SWE-bench MultilingualKimi K2.6 by 3.77376.7+3.7
- SWE-bench VerifiedKimi K2.6 by 3.476.880.2+3.4
- SciCodeKimi K2.6 by 2.54951.5+2.5
AgentsKimi K2.6
- AA ApexAgentsKimi K2.6 by 1711.528.5+17
- BrowseCompKimi K2.6 by 11.474.986.3+11.4
- GDPValKimi K2.6 by 9.817.227+9.8
- OSWorld-VerifiedKimi K2.6 by 9.863.373.1+9.8
- τ-Bench V3 · BankingKimi K2.6 by 9.114.223.3+9.1
- MCP AtlasKimi K2.6 by 4.363.868.1+4.3
ReasoningKimi K2.6
- Humanity's Last ExamKimi K2.6 by 6.830.737.5+6.8
- CritPtKimi K2.6 by 4.93.18+4.9
- GPQA DiamondKimi K2.6 by 3.287.991.1+3.2
FactsKimi K2.6
- AA-Omniscience · Non-hallucinationKimi K2.6 by 25.234.359.5+25.2
- Vectara HHEM hallucination ratelower is betterKimi K2.6 by 3.414.210.8+3.4
- AA-Omniscience · AccuracyKimi K2.5 by 2.635.232.6+2.6
- SimpleQA Verifiedtie34.334.9tie
Images and chartsKimi K2.6
- CharXiv (RQ)Kimi K2.6 by 9.277.586.7+9.2
- MMMU-ProKimi K2.6 by 475.479.4+4
- LMArena · VisionKimi K2.6 by 13 rating points12691282+13 rating
MathKimi K2.6
- HMMT Feb. 2026Kimi K2.6 by 5.687.192.7+5.6
- IMOAnswerBenchKimi K2.6 by 4.281.886+4.2
- AIME 2026tie95.896.4tie
Other results34 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.
- BabyVisionKimi K2.6 by 3236.568.5+32
- baby_vision_with_pythonKimi K2.6 by 2840.568.5+28
- MCPMarkKimi K2.6 by 26.429.555.9+26.4
- ToolathlonKimi K2.6 by 22.227.850+22.2
- GDPval-AA v2Kimi K2.6 by 181 rating points10091190+181 rating
- Terminal-Bench 2.0Kimi K2.6 by 15.950.866.7+15.9
- Artificial Analysis Coding IndexKimi K2.6 by 1546.861.8+15
- AA-OmniscienceKimi K2.6 by 12.6-7.35.3+12.6
- frontiermath_tier_4_v1Kimi K2.6 by 10.44.214.6+10.4
- V* (w/ python)Kimi K2.6 by 1086.996.9+10
- MATH-VisionKimi K2.6 by 984.293.2+9
- BrowseComp (w/ Ctx)Kimi K2.6 by 8.374.983.2+8.3
- mathvision_with_pythonKimi K2.6 by 8.28593.2+8.2
- WideSearch (item-f1)Kimi K2.6 by 8.172.780.8+8.1
- charxiv_rq_with_pythonKimi K2.6 by 878.786.7+8
- BrowseComp (Agent Swarm)Kimi K2.6 by 7.978.486.3+7.9
- SWEBench Pro PublicKimi K2.6 by 7.950.758.6+7.9
- vectara_answer_rateKimi K2.6 by 7.592.299.7+7.5
- τ³-Bench BankingKimi K2.6 by 6.414.220.6+6.4
- DeepSearchQA (accuracy)Kimi K2.6 by 5.977.183+5.9
- DeepSearchQA (F1)Kimi K2.6 by 5.977.183+5.9
- OJBench (python)Kimi K2.6 by 5.954.760.6+5.9
- Global-MMLU-LiteKimi K2.6 by 4.48488.4+4.4
- HLE (with tools)Kimi K2.6 by 3.850.254+3.8
- AA IntelligenceKimi K2.6 by 3.523.527+3.5
- vectara_factual_consistencyKimi K2.6 by 3.485.889.2+3.4
- LiveBenchKimi K2.6 by 3.169.172.2+3.1
- mmmu_pro_with_pythonKimi K2.6 by 2.477.780.1+2.4
- WideSearchKimi K2.6 by 1.87980.8+1.8
- vectara_avg_summary_lengthtie112116.7tie
- AA Agentic Indextie21.722.1tie
- StrongREJECTtie99.599.8tie
- MMLU-Protie87.187.1tie
- τ²-Bench Telecom (AA run)tie95.995.9tie
Questions people ask
Which is better, Kimi K2.5 or Kimi K2.6?
Kimi K2.6 wins all eight areas we test: coding, agents, reasoning, facts, images and charts, math, long documents and following instructions. Kimi K2.5 wins none, but costs 27% less.
Which is better for coding?
Kimi K2.6. It wins 8 of the 8 coding tests both models report; Kimi K2.5 wins none.
Which is cheaper?
Kimi K2.5 costs $0.60 per million input tokens and $3.00 per million output tokens; Kimi K2.6 costs $0.95 and $4.00. That makes Kimi K2.5 about 27% 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 29 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 34 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.