Model comparison
Claude Haiku 4.5 vs Kimi K2.6
Head-to-head evidence from 4 shared benchmark results across 3 categories. Overall scores shown here use BenchLM's provisional ranking lane.
Verified leaderboard positions: Claude Haiku 4.5 unranked; Kimi K2.6 #13
Evidence parity. Claude Haiku 4.5 and Kimi K2.6 share 4 comparable benchmark results. 2 of 8 categories are comparable. 1 result is unique to Claude Haiku 4.5; 56 to Kimi K2.6.
Updated July 13, 2026- Shared results
- 4
- Claude Haiku 4.5 only
- 1
- Kimi K2.6 only
- 56
- Comparable categories
- 2 / 8
Pick Kimi K2.6 if you want the stronger benchmark profile. Claude Haiku 4.5 only becomes the better choice if coding is the priority or you would rather avoid the extra latency and token burn of a reasoning model.
Confidence note. This is a partial-evidence comparison with 4 shared benchmark results across 3 evidence categories; 2 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 54. 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 mathematics, where it averages 67.1 against 4.9. The single biggest benchmark swing on the page is FrontierMath v2 (Tiers 1-3), 5.903% to 38.966%. Claude Haiku 4.5 does hit back in coding, so the answer changes if that is the part of the workload you care about most.
Claude Haiku 4.5 is also the more expensive model on tokens at $1.00 input / $5.00 output per 1M tokens, versus $0.95 input / $4.00 output per 1M tokens for Kimi K2.6. Kimi K2.6 is the reasoning model in the pair, while Claude Haiku 4.5 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. Kimi K2.6 gives you the larger context window at 256K, compared with 200K for Claude Haiku 4.5.
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 | Claude Haiku 4.5 | Δ | Kimi K2.6 |
|---|---|---|---|
| Math | Claude Haiku 4.54.9 | Margin→ 62.2 | Kimi K2.667.1 |
| Coding | Claude Haiku 4.573.3 | Margin← 0.7 | Kimi K2.672.6 |
| Agentic | Claude Haiku 4.5Not measured | MarginNo overlap | Kimi K2.673.5 |
| Knowledge | Claude Haiku 4.5Not measured | MarginNo overlap | Kimi K2.642.2 |
| Multimodal | Claude Haiku 4.5Not measured | MarginNo overlap | Kimi K2.679.8 |
Decisive benchmark drivers
The largest measured benchmark gaps in this matchup, with exact reported values.
More
- Source ↗
FrontierMath v2 (Tiers 1-3)
MathA 5.903%B 38.966%Winner: Kimi K2.6Δ 33.1FrontierMath v2 (Tiers 1-3): Claude Haiku 4.5 scored 5.903%; Kimi K2.6 scored 38.966%. Kimi K2.6 wins this benchmark. - Source ↗
FrontierMath v2 (Tier 4)
MathA 2.083%B 14.580%Winner: Kimi K2.6Δ 12.5FrontierMath v2 (Tier 4): Claude Haiku 4.5 scored 2.083%; Kimi K2.6 scored 14.580%. Kimi K2.6 wins this benchmark. - Source ↗
SWE-bench Verified
CodingA 73.3%B 80.2%Winner: Kimi K2.6Δ 6.9SWE-bench Verified: Claude Haiku 4.5 scored 73.3%; Kimi K2.6 scored 80.2%. Kimi K2.6 wins this benchmark.
Operational comparison
Runtime and commercial metrics are compared only when both models have a complete sourced value.
| Metric | Claude Haiku 4.5 | Kimi K2.6 | Comparison |
|---|---|---|---|
| Input / output priceUSD per 1M tokens | Claude Haiku 4.5$1 input / $5 output | Kimi K2.6$0.95 input / $4 output | Kimi K2.6 has the lower combined listed price. |
| Generation speedtokens per second | Claude Haiku 4.5Not available | Kimi K2.6Not available | A complete speed comparison is not available. |
| First-answer latencyseconds to first token | Claude Haiku 4.5Not available | Kimi K2.6Not available | A complete latency comparison is not available. |
| Context windowmaximum listed tokens | Claude Haiku 4.5200K | Kimi K2.6256K | Kimi K2.6 lists the larger context window. |
Benchmark Deep Dive
Agentic23 benchmarks
| Benchmark | Claude Haiku 4.5 | Kimi K2.6 | Result |
|---|---|---|---|
| JobBenchSource | 16.0% | — | Not comparable |
| Terminal-Bench 2.0Source | — | 66.7% | Not comparable |
| BrowseCompSource | — | 83.2% | Not comparable |
| OSWorld-VerifiedSource | — | 73.1% | Not comparable |
| ToolathlonSource | — | 50% | Not comparable |
| MCP AtlasSource | — | 55.9% | Not comparable |
| Claw-EvalSource | — | 62.3% | Not comparable |
| DeepSearchQASource | — | 92.5% | Not comparable |
| WideResearchSource | — | 80.8% | Not comparable |
| AA Agentic IndexSource | — | 30.3% | Not comparable |
| Tau2-TelecomSource | — | 95.9% | Not comparable |
| GDPval-AASource | — | 34.5% | Not comparable |
| GDPval-AASource | — | 1190 | Not comparable |
| APEX-Agents-AASource | — | 28.5% | Not comparable |
| Gert LabsSource | — | 56.82% | Not comparable |
| ResearchClawBenchSource | — | 18.0% | Not comparable |
| 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 |
CodingClaude Haiku 4.5 wins13 benchmarks
| Benchmark | Claude Haiku 4.5 | Kimi K2.6 | Result |
|---|---|---|---|
| SWE-bench VerifiedSource | 73.3% | 80.2% | Kimi K2.6 leads |
| LiveCodeBenchSource | — | 89.6% | Not comparable |
| LiveCodeBench v6Source | — | 89.6% | Not comparable |
| SWE-bench ProSource | — | 58.6% | Not comparable |
| SWE MultilingualSource | — | 76.7% | Not comparable |
| SciCodeSource | — | 52.2% | Not comparable |
| 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% | Not comparable |
| AA-SciCodeSource | — | 53.5% | Not comparable |
| AA Terminal-Bench 2.1Source | — | 65.9% | Not comparable |
Reasoning2 benchmarks
Knowledge10 benchmarks
| Benchmark | Claude Haiku 4.5 | Kimi K2.6 | Result |
|---|---|---|---|
| GPQASource | — | 90.5% | Not comparable |
| GPQA-DSource | — | 90.5% | Not comparable |
| HLESource | — | 34.7% | Not comparable |
| Artificial Analysis Intelligence IndexSource | — | 44.2% | Not comparable |
| AA-GPQA DiamondSource | — | 91.1% | Not comparable |
| AA-HLESource | — | 35.9% | Not comparable |
| AA-Omniscience IndexSource | — | 6.4% | Not comparable |
| AA-Omniscience AccuracySource | — | 32.8% | Not comparable |
| AA-Omniscience Hallucination RateSource | — | 39.3% | Not comparable |
| AA Openness IndexSource | — | 33.3% | Not comparable |
MathKimi K2.6 wins5 benchmarks
Multimodal7 benchmarks
| Benchmark | Claude Haiku 4.5 | Kimi K2.6 | Result |
|---|---|---|---|
| Design Arena WebsiteSource | 1164 | 1318 | Kimi K2.6 leads |
| MMMU-ProSource | — | 79.4% | Not comparable |
| 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% | Not comparable |
Inst. Following1 benchmarks
| Benchmark | Claude Haiku 4.5 | Kimi K2.6 | Result |
|---|---|---|---|
| AA-IFBenchSource | — | 76.0% | Not comparable |
Frequently Asked Questions (3)
Which is better, Claude Haiku 4.5 or Kimi K2.6?
Kimi K2.6 is ahead on BenchLM's provisional leaderboard, 74 to 54. The biggest single separator in this matchup is FrontierMath v2 (Tiers 1-3), where the scores are 5.903% and 38.966%.
Which is better for coding, Claude Haiku 4.5 or Kimi K2.6?
Claude Haiku 4.5 has the edge for coding in this comparison, averaging 73.3 versus 72.6. Inside this category, SWE-bench Verified is the benchmark that creates the most daylight between them.
Which is better for math, Claude Haiku 4.5 or Kimi K2.6?
Kimi K2.6 has the edge for math in this comparison, averaging 67.1 versus 4.9. Inside this category, FrontierMath v2 (Tiers 1-3) 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.