Kimi K2.6 vs Qwen3.8 27B
Wins 2 of 8 areas
Reasoning · Images and charts
Wins 4 of 8 areas
Agents · Math · Long documents · Following instructions
Qwen3.8 27B wins more areas, narrowly.Kimi K2.6 is better at reasoning.
Scores updated · 37 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 own15Qwen3.8 27B5 of 6 tests
- Following instructionsDoing exactly what it is asked02Qwen3.8 27B2 of 2 tests
- MathCompetition and research-level math01Qwen3.8 27B1 of 1 test
- Long documentsFinding answers in very long texts01Qwen3.8 27B1 of 1 test
- ReasoningHard problems that need careful thinking20Kimi K2.62 of 4 tests · 2 ties
- Images and chartsUnderstanding pictures, charts and video21Kimi K2.62 of 3 tests
- CodingWriting and fixing software44Even4 each · 1 tie
- FactsGetting facts right instead of making them up11Even1 each
Math and long documents 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.
Qwen3.8 27B costs 29% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Qwen3.8 27B pulls ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+24.7points ahead
- Real work tasks from 44 professionsGDPVal+19.1points ahead
- Fixes real GitHub issues, working as an agentLiveBench · Agentic Coding+14.5points ahead
Where Kimi K2.6 pulls ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+24.7points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+17points ahead
- Code for real scientific research problemsSciCode+4.9points ahead
Every test, side by side
All 37 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingEven
- LiveBench · Agentic CodingQwen3.8 27B by 14.546.961.4+14.5
- Terminal-Bench 2.1Qwen3.8 27B by 13.965.979.8+13.9
- LMArena · WebDevQwen3.8 27B by 81 rating points15091590+81 rating
- SciCodeKimi K2.6 by 4.951.546.6+4.9
- SWE-bench ProQwen3.8 27B by 3.158.661.7+3.1
- LiveBench · CodingKimi K2.6 by 2.978.675.7+2.9
- SWE-bench MultilingualKimi K2.6 by 2.976.773.8+2.9
- SWE-bench VerifiedKimi K2.6 by 2.480.277.8+2.4
- LiveCodeBench v6tie89.690.3tie
AgentsQwen3.8 27B
- BrowseCompKimi K2.6 by 24.786.361.6+24.7
- τ-Bench V3 · BankingQwen3.8 27B by 24.723.348+24.7
- GDPValQwen3.8 27B by 19.126.946+19.1
- OSWorld-VerifiedQwen3.8 27B by 11.273.184.3+11.2
- MCP AtlasQwen3.8 27B by 9.268.177.3+9.2
- Terminal-Bench 4.0Qwen3.8 27B by 5.10.55.6+5.1
ReasoningKimi K2.6
- Humanity's Last ExamKimi K2.6 by 3.637.533.9+3.6
- CritPtKimi K2.6 by 2.685.4+2.6
- GPQA Diamondtie91.190.5tie
- LiveBench · Reasoningtie79.480tie
FactsEven
- AA-Omniscience · AccuracyKimi K2.6 by 1732.615.6+17
- AA-Omniscience · Non-hallucinationQwen3.8 27B by 10.259.569.7+10.2
Images and chartsKimi K2.6
- CharXiv (RQ)Qwen3.8 27B by 3.586.790.2+3.5
- MMMU-ProKimi K2.6 by 3.179.476.3+3.1
- LMArena · VisionKimi K2.6 by 11 rating points12821271+11 rating
Following instructionsQwen3.8 27B
- LiveBench · Instruction FollowingQwen3.8 27B by 8.364.472.7+8.3
- IFBenchQwen3.8 27B by 3.57679.5+3.5
Other results9 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 Agentic IndexQwen3.8 27B by 24.422.146.5+24.4
- BabyVisionQwen3.8 27B by 17.168.585.6+17.1
- AA-OmniscienceKimi K2.6 by 15.35.3-10+15.3
- livebench_data_analysisQwen3.8 27B by 11.565.176.6+11.5
- AA IntelligenceQwen3.8 27B by 6.72733.7+6.7
- Artificial Analysis Coding IndexQwen3.8 27B by 6.361.868.1+6.3
- DeepSearchQA (F1)Kimi K2.6 by 3.38379.7+3.3
- MATH-VisionQwen3.8 27B by 1.493.294.6+1.4
- livebench_languagetie75.174.3tie
Questions people ask
Which is better, Kimi K2.6 or Qwen3.8 27B?
Qwen3.8 27B wins four of the eight areas we test: agents, math, long documents and following instructions. Kimi K2.6 wins reasoning and images and charts. They are level on coding and facts.
Which is better for coding?
Neither. They win 4 coding tests each of the 9 both models report, and 1 is a tie.
Which is cheaper?
Kimi K2.6 costs $0.95 per million input tokens and $4.00 per million output tokens; Qwen3.8 27B costs $0.50 and $3.00. That makes Qwen3.8 27B about 29% cheaper for the same work.
How do you compare the two?
We use the 37 benchmark tests both models have published scores on. The verdict counts the 28 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 9 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.