Compare / Kimi K2.5 vs Kimi K2.6

Kimi K2.5 vs Kimi K2.6

Which is cheaper, which is more capable, and which does better at each kind of work, from 31 results on 7 sources that measured both models.

Moonshot AIKimi K2.5 Released 27 Jan 2026 · $0.57 in · $2.85 out
Moonshot AIKimi K2.6 Released 20 Apr 2026 · $0.95 in · $4.00 out
Kimi K2.5 wins 13 too close to callKimi K2.6 wins 27

Of the 31 results both models have, the better value on each. Results whose reported ranges overlap are too close to call.

Results as of 5 November 2026 (catalogue release 2026-11-05-44c5fb63d296).

Kimi K2.5 or Kimi K2.6? The short answer

Kimi K2.6 is the newer version and scores higher (Intelligence Index 27.0 against 23.5), but Kimi K2.5 is cheaper (33% less per token): upgrade if the gain is worth the price.

  • Kimi K2.5 is cheaper: $1.14 against $1.71 per million tokens, 33% less.
  • Kimi K2.6 scores higher on the Artificial Analysis Intelligence Index: 27.0 against 23.5.
  • Kimi K2.6 is newer: released 20 Apr 2026, against 27 Jan 2026 for Kimi K2.5.
  • On third-party benchmarks, on the results both models have: Kimi K2.6 does better for agents and tool use, reasoning and knowledge, writing and sticking to the facts.

Head to head

23.5
Intelligence Index
27.0
$1.14
Price per million tokens, blendedlower is better
$1.71
262,144
Context window, tokens
262,144

Every fact side by side

Kimi K2.5Kimi K2.6
Price per million tokens$0.57 in · $2.85 out$0.95 in · $4.00 out
Blended (3 in : 1 out)$1.14$1.71
Cheapest host, blended$0.99 (AtlasCloud, INT4)$0.70 (Baidu, FP4)
Cached input, per million–$0.16 cached input
Intelligence Index23.527.0
Spring Prompt overall––
Output speed, tokens a second––
Context window262,144 tokens262,144 tokens
Released27 Jan 202620 Apr 2026 (newer)
WeightsOpenOpen
Developer based inChinaChina

Prices are the developer's list price where we have read it, otherwise the typical host on OpenRouter; see every model's price. Intelligence and speed from Artificial Analysis.

Which is better for what

Each model's rank among the models on sale today, for each kind of work, from our own benchmarks and licensed sources. Further right is better; the better rank is in green.

Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls 92 of 160 66 of 160
Professional workReal tasks from banking, consulting and law, business documents and freelance projects not measured 43 of 51
Reasoning and knowledgeHard questions across science, maths and general knowledge 75 of 188 54 of 188
WritingWhat people prefer in blind comparisons, and judged writing quality 47 of 173 36 of 173
Sticking to the factsSummarising without inventing things, and factual answers 101 of 143 79 of 143

Every shared result

Each model's best result on every measure both have, grouped by benchmark, with the setting that produced it. Under each measure, every model that benchmark has measured is a grey tick, best to the right, so you can see where the two sit in the field. The headline measures of each benchmark come first; the rest are under "Show all". Green marks the better value; where a source reports ranges that overlap, neither is marked.

MeasureKimi K2.5Kimi K2.6
Arena (formerly LMArena) reported by Arena (formerly LMArena) · 3 results
Overall · TextArena rating, higher is better 67 · 50 of 387 measured 1,450thinking reasoning 1,461
Overall · Text factualityArena rating, higher is better 69 · 45 of 163 measured 1,445thinking reasoning 1,456
Business, management and finance · TextArena rating, higher is better 86 · 59 of 380 measured 1,440thinking reasoning 1,455
Artificial Analysis reported by Artificial Analysis · 3 results
Artificial Analysis Intelligence Indexindex score, higher is better 88 · 66 of 249 measured 23.5 27.0
GPQA Diamond% of questions, higher is better 60 · 31 of 232 measured 87.9% 91.1%
Humanity's Last Exam% of questions, higher is better 76 · 55 of 248 measured 30.7% 37.5%
OpenHands Index reported by OpenHands Index · 3 results
Average% resolved, higher is better 18 · 14 of 34 measured 49.2%openhands 57.1%openhands
Front end (SWE-Bench Multimodal)% resolved, higher is better 18 · 6 of 34 measured 32.8%openhands 41.2%openhands
Greenfield (Commit0)% resolved, higher is better 21 · 14 of 34 measured 18.8%openhands 25.0%openhands
SimpleQA Verified (Epoch AI) reported by SimpleQA Verified (Epoch AI) · 1 result
Correct answers% of questions, higher is better 50 · 49 of 76 measured 34.3% 34.9%
UGI Leaderboard reported by UGI Leaderboard · 3 results
Writing score · UGIscore out of 100, higher is better 38 · 30 of 230 measured 62.4thinking reasoning 64.4thinking reasoning
Requested-length error · UGI% off the requested word count, lower is better 43 · 91 of 230 measured 10.0%thinking reasoning 18.0%thinking reasoning
Style adherence · UGIscore from 0 to 1, higher is better 126 · 108 of 230 measured 0.34no reasoning 0.35thinking reasoning
Vectara Hallucination Leaderboard reported by Vectara Hallucination Leaderboard · 2 results
Hallucination rate · Vectara% of summaries, lower is better 88 · 68 of 104 measured 14.2% 10.8%
Answer rate · Vectara% of documents, higher is better 97 · 41 of 104 measured 92.2% 99.7%
Vending-Bench 2 reported by Vending-Bench 2 · 1 result
Money after a yearUS dollars, higher is better 42 · 15 of 58 measured $1,198 $6,205
Show all 31 results (15 more)
MeasureKimi K2.5Kimi K2.6
Arena (formerly LMArena) reported by Arena (formerly LMArena) · 5 results
Creative writing · TextArena rating, higher is better 67 · 58 of 385 measured 1,423thinking reasoning 1,433
Expert prompts · TextArena rating, higher is better 58 · 32 of 338 measured 1,483thinking reasoning 1,503
Instruction following · TextArena rating, higher is better 69 · 47 of 387 measured 1,439thinking reasoning 1,456
Overall · DocumentArena rating, higher is better 32 · 27 of 38 measured 1,436thinking reasoning 1,450
Writing, literature and language · TextArena rating, higher is better 70 · 58 of 386 measured 1,428thinking reasoning 1,441
Artificial Analysis reported by Artificial Analysis · 7 results
Terminal-Bench 2.1% of tasks, higher is better 67 · 39 of 131 measured 45.7% 65.9%
τ-bench banking% of tasks, higher is better 62 · 42 of 121 measured 14.2% 23.3%
Artificial Analysis Coding Indexindex score, higher is better 62 · 37 of 130 measured 46.8 61.8
IFBench% of instructions, higher is better 54 · 18 of 195 measured 70.2% 76.0%
Long-context reasoning (AA-LCR)% of questions, higher is better 63 · 32 of 239 measured 78.0% 81.0%
Terminal-Bench Hard% of tasks, higher is better 49 · 23 of 190 measured 34.8% 43.9%
τ²-bench telecom% of tasks, higher is better 12 · 12 of 193 measured 95.9% 95.9%
OpenHands Index reported by OpenHands Index · 3 results
Information gathering (GAIA)% resolved, higher is better 17 · 9 of 34 measured 63.6%openhands 74.5%openhands
Issue resolution (SWE-Bench)% resolved, higher is better 27 · 13 of 34 measured 68.8%openhands 74.6%openhands
Testing (SWT-Bench)% resolved, higher is better 22 · 12 of 34 measured 61.9%openhands 70.4%openhands

Quick answers

Is Kimi K2.5 better than Kimi K2.6?

Kimi K2.6 is the newer version and scores higher (Intelligence Index 27.0 against 23.5), but Kimi K2.5 is cheaper (33% less per token): upgrade if the gain is worth the price.

Which is cheaper, Kimi K2.5 or Kimi K2.6?

Kimi K2.5 is cheaper: $1.14 against $1.71 per million tokens, 33% less.

Which is better for agents and tool use, Kimi K2.5 or Kimi K2.6?

Kimi K2.6 ranks 66 of 160 models on sale for multi-step tasks with tools: support desks, coding agents, function calls, against 92 for Kimi K2.5.

Which is better for reasoning and knowledge, Kimi K2.5 or Kimi K2.6?

Kimi K2.6 ranks 54 of 188 models on sale for hard questions across science, maths and general knowledge, against 75 for Kimi K2.5.

Which is better for writing, Kimi K2.5 or Kimi K2.6?

Kimi K2.6 ranks 36 of 173 models on sale for what people prefer in blind comparisons, and judged writing quality, against 47 for Kimi K2.5.

Which is better for sticking to the facts, Kimi K2.5 or Kimi K2.6?

Kimi K2.6 ranks 79 of 143 models on sale for summarising without inventing things, and factual answers, against 101 for Kimi K2.5.

Run Kimi K2.5 and Kimi K2.6 on your own prompt.

Benchmarks aren't your task. Test both side by side on your prompt, with every answer checked against your rules and the cost of each one.