Kimi K2.5 vs Qwen3.5 397B A17B
Wins 5 of 8 areas
Coding · Agents · Reasoning · Facts · Math
Wins 3 of 8 areas
Images and charts · Long documents · Following instructions
Kimi K2.5 is the stronger all-rounder.Qwen3.5 397B A17B is better at following instructions.
Scores updated · 48 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.
- FactsGetting facts right instead of making them up30Kimi K2.53 of 3 tests
- CodingWriting and fixing software42Kimi K2.54 of 8 tests · 2 ties
- AgentsCarrying out multi-step tasks on its own21Kimi K2.52 of 4 tests · 1 tie
- ReasoningHard problems that need careful thinking21Kimi K2.52 of 3 tests
- MathCompetition and research-level math10Kimi K2.51 of 3 tests · 2 ties
- Following instructionsDoing exactly what it is asked02Qwen3.5 397B A17B2 of 2 tests
- Images and chartsUnderstanding pictures, charts and video01Qwen3.5 397B A17B1 of 2 tests · 1 tie
- Long documentsFinding answers in very long texts01Qwen3.5 397B A17B1 of 2 tests · 1 tie
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 14% 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 K2.5 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+23.2points ahead
- Short factual questions, answered correctlySimpleQA Verified+8.3points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+5.9points ahead
Where Qwen3.5 397B A17B pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+8.6points ahead
- Keeps track of context across a multi-turn chatMulti-Challenge+6.2points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+6.1points ahead
Every test, side by side
All 48 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingKimi K2.5
- Terminal-Bench HardQwen3.5 397B A17B by 6.134.840.9+6.1
- Terminal-Bench 2.1Qwen3.5 397B A17B by 5.645.751.3+5.6
- SciCodeKimi K2.5 by 4.24944.8+4.2
- SWE-bench MultilingualKimi K2.5 by 3.77369.3+3.7
- LMArena · WebDevKimi K2.5 by 36 rating points14361400+36 rating
- LiveCodeBench v6Kimi K2.5 by 1.48583.6+1.4
- SWE-bench Verifiedtie76.876.4tie
- SWE-bench Protie50.750.9tie
AgentsKimi K2.5
- BrowseCompKimi K2.5 by 5.974.969+5.9
- AA ApexAgentsQwen3.5 397B A17B by 3.811.515.3+3.8
- GDPValKimi K2.5 by 2.417.214.8+2.4
- τ-Bench V3 · Bankingtie14.213.4tie
ReasoningKimi K2.5
- Humanity's Last ExamKimi K2.5 by 1.730.729+1.7
- CritPtKimi K2.5 by 1.43.11.7+1.4
- GPQA DiamondQwen3.5 397B A17B by 1.487.989.3+1.4
FactsKimi K2.5
- AA-Omniscience · Non-hallucinationKimi K2.5 by 23.234.311.1+23.2
- SimpleQA VerifiedKimi K2.5 by 8.334.326+8.3
- AA-Omniscience · AccuracyKimi K2.5 by 4.435.230.8+4.4
Images and chartsQwen3.5 397B A17B
- MMMU-ProQwen3.5 397B A17B by 1.975.477.3+1.9
- LMArena · Visiontie12691264tie
MathKimi K2.5
- AIME 2026Kimi K2.5 by 4.595.891.3+4.5
- IMOAnswerBenchtie81.880.9tie
- HMMT Feb. 2026tie87.187.9tie
Long documentsQwen3.5 397B A17B
- LongBench v2Qwen3.5 397B A17B by 2.26163.2+2.2
- AA-LCRtie7877.3tie
Following instructionsQwen3.5 397B A17B
- IFBenchQwen3.5 397B A17B by 8.670.278.8+8.6
- Multi-ChallengeQwen3.5 397B A17B by 6.261.467.6+6.2
Other results21 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-OmniscienceKimi K2.5 by 23.4-7.3-30.8+23.4
- MCPMarkQwen3.5 397B A17B by 16.629.546.1+16.6
- AA Agentic IndexKimi K2.5 by 11.121.710.6+11.1
- Seal-0Kimi K2.5 by 10.557.446.9+10.5
- ToolathlonQwen3.5 397B A17B by 10.527.838.3+10.5
- BrowseComp-ZHQwen3.5 397B A17B by 862.370.3+8
- AA IntelligenceKimi K2.5 by 5.123.518.4+5.1
- WideSearchKimi K2.5 by 57974+5
- GDPval-AA v2Kimi K2.5 by 47 rating points1009962+47 rating
- HMMT Nov. 2025Qwen3.5 397B A17B by 3.589.292.7+3.5
- HMMT 2025Kimi K2.5 by 2.795.492.7+2.7
- HLE (with tools)Kimi K2.5 by 1.950.248.3+1.9
- Terminal-Bench 2.0Qwen3.5 397B A17B by 1.750.852.5+1.7
- Artificial Analysis Coding IndexQwen3.5 397B A17B by 1.446.848.2+1.4
- τ³-Bench Bankingtie14.213.4tie
- MMLU-Protie87.187.8tie
- HMMT Feb. 2025tie95.494.8tie
- τ²-Bench Telecom (AA run)tie95.995.6tie
- NL2Repotie3232.2tie
- SWEBench Pro Publictie50.750.9tie
- StrongREJECTtie99.599.4tie
Questions people ask
Which is better, Kimi K2.5 or Qwen3.5 397B A17B?
Kimi K2.5 wins five of the eight areas we test: coding, agents, reasoning, facts and math. Qwen3.5 397B A17B wins images and charts, long documents and following instructions.
Which is better for coding?
Kimi K2.5. It wins 4 of the 8 coding tests both models report; Qwen3.5 397B A17B wins 2, and 2 are ties.
Which is cheaper?
Kimi K2.5 costs $0.60 per million input tokens and $3.00 per million output tokens; Qwen3.5 397B A17B costs $0.60 and $3.60. That makes Kimi K2.5 about 14% cheaper for the same work.
How do you compare the two?
We use the 48 benchmark tests both models have published scores on. The verdict counts the 27 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 21 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.