Kimi K2.5 kimi-k2.5
moonshotai · kimi reasoning flagship
#9 overall
- params
- 1059B (A32B)
- arch
- moe
- context
- 256k
- license
- modified-mit open weights
- released
- Feb 2026
- reasoning
- yes
- train compute
- 5.8e+24 FLOP
- downloads/30d
- 1.3M
# why ranked
Overall: #9 — score 43.5 ● medium — one signal missing, weights renormalized
| signal | weight | input (0–100) |
|---|---|---|
| aa_intelligence | 0.50 | 38.1 |
| arena_elo | 0.30 | — |
| bench_composite | 0.20 | 56.9 |
missing signals are dropped and the remaining weights renormalized — never imputed.
# trend
collecting history — trends appear after a few daily runs.
# benchmarks
# where to run
| provider | quant | ctx | $/M in | $/M out | $/M blend | tps | uptime |
|---|---|---|---|---|---|---|---|
| ModelRun | fp4 | 256k | $0.40 | $1.90 | $0.78 | — | 100.0% |
| DigitalOcean | unknown | 250k | $0.38 | $2.02 | $0.79 | — | 99.1% |
| Chutes | int4 | 256k | $0.44 | $2.00 | $0.83 | — | 83.9% |
| DeepInfra | fp4 | 256k | $0.45 | $2.25 | $0.90 | — | 99.7% |
| SiliconFlow | int4 | 256k | $0.45 | $2.25 | $0.90 | — | 100.0% |
| AtlasCloud | int4 | 256k | $0.49 | $2.50 | $0.99 | — | 100.0% |
| StreamLake | fp8 | 250k | $0.54 | $2.70 | $1.08 | — | 99.9% |
| Novita | unknown | 256k | $0.57 | $2.85 | $1.14 | — | 100.0% |
| Moonshot AI | int4 | 256k | $0.60 | $3.00 | $1.20 | — | 100.0% |
| Phala | unknown | 256k | $0.60 | $3.00 | $1.20 | — | — |
| Venice | unknown | 250k | $0.56 | $3.50 | $1.29 | — | — |
blended = (3·input + output) / 4 per 1M tokens · cheapest first
source aliases
- aa
kimi-k2.5- epoch
Kimi K2.5- llmstats
kimi-k2.5- openrouter
moonshotai/kimi-k2.5