Kimi K2.5
Language AvailableCompare models
ComparingKimi K2.5with
Pick a model above to see the comparison.
moonshot/kimi-k2.5 · by Moonshot AI · mixture-of-experts
Pricing — 1 offering(s)
Input tokens
- $0.60 / 1M tokens (input) Current 2026-01-27 → present
Output tokens
- $3.00 / 1M tokens (output) Current 2026-01-27 → present
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
Capability profile
coding strong
context window strong
reasoning unknown
tool use unknown
instruction following unknown
multilingual unknown
speed unknown
cost efficiency unknown
Coding benchmarks
View full leaderboard →| Benchmark | Score | Harness | Source |
|---|---|---|---|
| swe-bench-verified ↗ | 70.0% 2026-01 | mini-swe-agent | 📊 leaderboard ↗ |
Operator guidance
Two generations behind Moonshot's current flagship (K3) but still status: ga and independently confirmed on SWE-bench Verified. Prefer K2.6 or K3 for new integrations; this entry exists primarily because Cursor's Composer 2.5 was built by post-training this exact checkpoint (see that model's derived_from in modelglass-coding).
Use cases
- General agentic, coding, and visual-understanding tasks (open-source SoTA at release per Moonshot's own claim)
Limitations
- Capability ratings beyond coding/context-window are unknown — no independent benchmark found for reasoning/tool-use/multilingual/speed/cost-efficiency as of this backfill
- Status corrected during this backfill from an earlier 'retired' assumption (SCO-293) to ga — see the companion registry file's notes for the full correction