Model comparison
Kimi K2.6 vs Kimi K2.5
Head-to-head evidence from 39 shared benchmark results across 7 categories. Overall scores shown here use BenchLM's provisional ranking lane.
Verified leaderboard positions: Kimi K2.6 #13; Kimi K2.5 #27
Evidence parity. Kimi K2.6 and Kimi K2.5 share 39 comparable benchmark results. 5 of 8 categories are comparable. 21 results are unique to Kimi K2.6; 22 to Kimi K2.5.
Updated July 13, 2026- Shared results
- 39
- Kimi K2.6 only
- 21
- Kimi K2.5 only
- 22
- Comparable categories
- 5 / 8
Pick Kimi K2.6 if you want the stronger benchmark profile. Kimi K2.5 only becomes the better choice if knowledge is the priority or you want the cheaper token bill.
Confidence note. This is a partial-evidence comparison with 39 shared benchmark results across 7 evidence categories; 5 of 8 categories currently have scoreable aggregates for both models. Treat the verdict as directional until coverage is more balanced.
Why this result
Kimi K2.6 is clearly ahead on the provisional aggregate, 74 to 63. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
Kimi K2.6's sharpest advantage is in agentic, where it averages 73.5 against 55. The single biggest benchmark swing on the page is BrowseComp, 83.2% to 60.6%. Kimi K2.5 does hit back in knowledge, so the answer changes if that is the part of the workload you care about most.
Kimi K2.6 is also the more expensive model on tokens at $0.95 input / $4.00 output per 1M tokens, versus $0.60 input / $3.00 output per 1M tokens for Kimi K2.5. Kimi K2.6 is the reasoning model in the pair, while Kimi K2.5 is not. That usually helps on harder chain-of-thought-heavy tests, but it can also mean more latency and more token spend in real use.
Category breakdown
Exact category averages are shown below. Not measured means BenchLM does not have enough sourced public coverage for that model and category.
| Category | Kimi K2.6 | Δ | Kimi K2.5 |
|---|---|---|---|
| Agentic | Kimi K2.673.5 | Margin← 18.5 | Kimi K2.555.0 |
| Knowledge | Kimi K2.642.2 | Margin→ 15.0 | Kimi K2.557.2 |
| Coding | Kimi K2.672.6 | Margin← 7.1 | Kimi K2.565.5 |
| Math | Kimi K2.667.1 | Margin← 6.5 | Kimi K2.560.6 |
| Multimodal | Kimi K2.679.8 | Margin← 1.3 | Kimi K2.578.5 |
| Reasoning | Kimi K2.6Not measured | MarginNo overlap | Kimi K2.561.0 |
| Multilingual | Kimi K2.6Not measured | MarginNo overlap | Kimi K2.582.3 |
| Inst. Following | Kimi K2.6Not measured | MarginNo overlap | Kimi K2.593.9 |
Decisive benchmark drivers
The largest measured benchmark gaps in this matchup, with exact reported values.
More
- Source ↗
BrowseComp
AgenticA 83.2%B 60.6%Winner: Kimi K2.6Δ 22.6BrowseComp: Kimi K2.6 scored 83.2%; Kimi K2.5 scored 60.6%. Kimi K2.6 wins this benchmark. - Source ↗
Terminal-Bench 2.0
AgenticA 66.7%B 50.8%Winner: Kimi K2.6Δ 15.9Terminal-Bench 2.0: Kimi K2.6 scored 66.7%; Kimi K2.5 scored 50.8%. Kimi K2.6 wins this benchmark. - Source ↗
FrontierMath v2 (Tiers 1-3)
MathA 38.966%B 27.900%Winner: Kimi K2.6Δ 11.1FrontierMath v2 (Tiers 1-3): Kimi K2.6 scored 38.966%; Kimi K2.5 scored 27.900%. Kimi K2.6 wins this benchmark. - Source ↗
FrontierMath v2 (Tier 4)
MathA 14.580%B 4.200%Winner: Kimi K2.6Δ 10.4FrontierMath v2 (Tier 4): Kimi K2.6 scored 14.580%; Kimi K2.5 scored 4.200%. Kimi K2.6 wins this benchmark. - Source ↗
SWE-bench Pro
CodingA 58.6%B 50.7%Winner: Kimi K2.6Δ 7.9SWE-bench Pro: Kimi K2.6 scored 58.6%; Kimi K2.5 scored 50.7%. Kimi K2.6 wins this benchmark.
Operational comparison
Runtime and commercial metrics are compared only when both models have a complete sourced value.
| Metric | Kimi K2.6 | Kimi K2.5 | Comparison |
|---|---|---|---|
| Input / output priceUSD per 1M tokens | Kimi K2.6$0.95 input / $4 output | Kimi K2.5$0.6 input / $3 output | Kimi K2.5 has the lower combined listed price. |
| Generation speedtokens per second | Kimi K2.6Not available | Kimi K2.545 tok/s | A complete speed comparison is not available. |
| First-answer latencyseconds to first token | Kimi K2.6Not available | Kimi K2.52.38 s | A complete latency comparison is not available. |
| Context windowmaximum listed tokens | Kimi K2.6256K | Kimi K2.5256K | Listed context windows are equal. |
Benchmark Deep Dive
AgenticKimi K2.6 wins27 benchmarks
| Benchmark | Kimi K2.6 | Kimi K2.5 | Result |
|---|---|---|---|
| Terminal-Bench 2.0Source | 66.7% | 50.8% | Kimi K2.6 leads |
| BrowseCompSource | 83.2% | 60.6% | Kimi K2.6 leads |
| OSWorld-VerifiedSource | 73.1% | — | Not comparable |
| ToolathlonSource | 50% | 27.8% | Kimi K2.6 leads |
| MCP AtlasSource | 55.9% | 29.5% | Kimi K2.6 leads |
| Claw-EvalSource | 62.3% | 52.3% | Kimi K2.6 leads |
| DeepSearchQASource | 92.5% | 77.1% | Kimi K2.6 leads |
| WideResearchSource | 80.8% | 72.7% | Kimi K2.6 leads |
| AA Agentic IndexSource | 30.3% | — | Not comparable |
| Tau2-TelecomSource | 95.9% | 95.9% | Tie |
| GDPval-AASource | 34.5% | — | Not comparable |
| GDPval-AASource | 1190 | — | Not comparable |
| APEX-Agents-AASource | 28.5% | 11.5% | Kimi K2.6 leads |
| Gert LabsSource | 56.82% | 45.88% | Kimi K2.6 leads |
| ResearchClawBenchSource | 18.0% | 14.0% | Kimi K2.6 leads |
| OSWorld 2.0Source | 4.6% | — | Not comparable |
| AA BriefcaseSource | 809 | — | Not comparable |
| AA AutomationBenchSource | 19.6% | — | Not comparable |
| AA EnterpriseOps-GymSource | 38.5% | — | Not comparable |
| AA Harvey LABSource | 0.0% | — | Not comparable |
| AA ITBenchSource | 31.2% | — | Not comparable |
| AA Tau3 BankingSource | 20.6% | — | Not comparable |
| QwenClawBenchSource | — | 54.3% | Not comparable |
| TAU3-BenchSource | — | 65.7% | Not comparable |
| DeepPlanningSource | — | 14.4% | Not comparable |
| MCP-TasksSource | — | 59.1% | Not comparable |
| JobBenchSource | — | 8.7% | Not comparable |
CodingKimi K2.6 wins16 benchmarks
| Benchmark | Kimi K2.6 | Kimi K2.5 | Result |
|---|---|---|---|
| SWE-bench VerifiedSource | 80.2% | 76.8% | Kimi K2.6 leads |
| LiveCodeBenchSource | 89.6% | 85% | Kimi K2.6 leads |
| LiveCodeBench v6Source | 89.6% | 85.0% | Kimi K2.6 leads |
| SWE-bench ProSource | 58.6% | 50.7% | Kimi K2.6 leads |
| SWE MultilingualSource | 76.7% | 73% | Kimi K2.6 leads |
| SciCodeSource | 52.2% | 48.7% | Kimi K2.6 leads |
| Terminal-Bench 2.0Source | 66.7% | — | Not comparable |
| Vibe Code BenchSource | 37.89% | — | Not comparable |
| cursorBench31Source | 47.6% | — | Not comparable |
| AA Coding IndexSource | 61.8% | — | Not comparable |
| Terminal-Bench HardSource | 43.9% | 34.8% | Kimi K2.6 leads |
| AA-SciCodeSource | 53.5% | 49.0% | Kimi K2.6 leads |
| AA Terminal-Bench 2.1Source | 65.9% | — | Not comparable |
| SWE-bench Verified*Source | — | 70.8% | Not comparable |
| SWE-RebenchSource | — | 58.5% | Not comparable |
| React Native EvalsSource | — | 77.2% | Not comparable |
Reasoning3 benchmarks
KnowledgeKimi K2.5 wins13 benchmarks
| Benchmark | Kimi K2.6 | Kimi K2.5 | Result |
|---|---|---|---|
| GPQASource | 90.5% | 87.6% | Kimi K2.6 leads |
| GPQA-DSource | 90.5% | 87.6% | Kimi K2.6 leads |
| HLESource | 34.7% | 30.1% | Kimi K2.6 leads |
| Artificial Analysis Intelligence IndexSource | 44.2% | 38.1% | Kimi K2.6 leads |
| AA-GPQA DiamondSource | 91.1% | 87.9% | Kimi K2.6 leads |
| AA-HLESource | 35.9% | 29.4% | Kimi K2.6 leads |
| AA-Omniscience IndexSource | 6.4% | -8.1% | Kimi K2.6 leads |
| AA-Omniscience AccuracySource | 32.8% | 34.3% | Kimi K2.5 leads |
| AA-Omniscience Hallucination RateSource | 39.3% | 64.6% | Kimi K2.6 leads |
| AA Openness IndexSource | 33.3% | — | Not comparable |
| SuperGPQASource | — | 69.2% | Not comparable |
| MMLU-ProSource | — | 87.1% | Not comparable |
| MMLU-Pro (Arcee)Source | — | 87.1% | Not comparable |
MathKimi K2.6 wins9 benchmarks
| Benchmark | Kimi K2.6 | Kimi K2.5 | Result |
|---|---|---|---|
| AIME26Source | 96.4% | 95.8% | Kimi K2.6 leads |
| HMMT Feb 2026Source | 92.7% | 87.1% | Kimi K2.6 leads |
| MMAnswerBenchSource | 86.0% | 81.8% | Kimi K2.6 leads |
| FrontierMath v2 (Tiers 1-3)Source | 38.966% | 27.900% | Kimi K2.6 leads |
| FrontierMath v2 (Tier 4)Source | 14.580% | 4.200% | Kimi K2.6 leads |
| AIME 2025Source | — | 96.1% | Not comparable |
| AIME25 (Arcee)Source | — | 96.3% | Not comparable |
| HMMT Feb 2025Source | — | 95.4% | Not comparable |
| HMMT Nov 2025Source | — | 91.1% | Not comparable |
Multilingual2 benchmarks
MultimodalKimi K2.6 wins10 benchmarks
| Benchmark | Kimi K2.6 | Kimi K2.5 | Result |
|---|---|---|---|
| MMMU-ProSource | 79.4% | 78.5% | Kimi K2.6 leads |
| MMMU-Pro w/ PythonSource | 80.1% | — | Not comparable |
| CharXivSource | 80.4% | — | Not comparable |
| MathVisionSource | 87.4% | — | Not comparable |
| V*Source | 96.9% | — | Not comparable |
| AA-MMMU-ProSource | 79.4% | 75.4% | Kimi K2.6 leads |
| Design Arena WebsiteSource | 1318 | 1291 | Kimi K2.6 leads |
| Video-MMESource | — | 87.4% | Not comparable |
| MMVUSource | — | 80.4% | Not comparable |
| VideoMMMUSource | — | 86.6% | Not comparable |
Frequently Asked Questions (6)
Which is better, Kimi K2.6 or Kimi K2.5?
Kimi K2.6 is ahead on BenchLM's provisional leaderboard, 74 to 63. The biggest single separator in this matchup is BrowseComp, where the scores are 83.2% and 60.6%.
Which is better for knowledge tasks, Kimi K2.6 or Kimi K2.5?
Kimi K2.5 has the edge for knowledge tasks in this comparison, averaging 57.2 versus 42.2. Inside this category, AA-Omniscience Hallucination Rate is the benchmark that creates the most daylight between them.
Which is better for coding, Kimi K2.6 or Kimi K2.5?
Kimi K2.6 has the edge for coding in this comparison, averaging 72.6 versus 65.5. Inside this category, Terminal-Bench Hard is the benchmark that creates the most daylight between them.
Which is better for math, Kimi K2.6 or Kimi K2.5?
Kimi K2.6 has the edge for math in this comparison, averaging 67.1 versus 60.6. Inside this category, FrontierMath v2 (Tiers 1-3) is the benchmark that creates the most daylight between them.
Which is better for agentic tasks, Kimi K2.6 or Kimi K2.5?
Kimi K2.6 has the edge for agentic tasks in this comparison, averaging 73.5 versus 55. Inside this category, MCP Atlas is the benchmark that creates the most daylight between them.
Which is better for multimodal and grounded tasks, Kimi K2.6 or Kimi K2.5?
Kimi K2.6 has the edge for multimodal and grounded tasks in this comparison, averaging 79.8 versus 78.5. Inside this category, Design Arena Website is the benchmark that creates the most daylight between them.
Self-host vs API cost
Estimates at 50,000 req/day · 1000 tokens/req average.
Related Comparisons
The AI models change fast. We track them for you.
A weekly brief for engineers and researchers covering new models, ranking shifts, and pricing changes.
Free. No spam. Unsubscribe anytime.