Kimi K2 Instruct 0905 kimi-k2-0905
moonshotai · kimi agentic
#22 overall #4 agentic
- params
- 1000B (A32B)
- arch
- moe
- context
- 256k
- license
- modified-mit open weights
- released
- Sep 2025
- downloads/30d
- 39.0k
# architecture
- layers
- 61
- d_model
- 7168
- heads
- 64q / 64kv
- head dim
- 112
- experts
- top-8 of 384 + 1 shared
- vocab
- 163840
- family
- kimi_k2
signal path of one layer, generated from the registry's structured fields — dimension lines quote the model's real numbers; an MoE trace forks at the router, a dense trace runs straight through. Geometry fields come from the repo's config.json.
# why ranked
Overall: #22 — score 65.5 ● high — all signals present
| signal | weight | input (0–100) |
|---|---|---|
| aa_intelligence | 0.70 | 56.7 |
| bench_composite | 0.30 | 85.9 |
benchmark panel evidence:
| complete panel | score (0–100) |
|---|---|
| swe-bench-verified-v1 | 85.9 |
Coding: provisional — score 88.9 ● low — a single signal; treat with caution
provisional: too few independent signals for a numbered position — the score is shown, but this model sorts after every ranked model.
| signal | weight | input (0–100) |
|---|---|---|
| aa_coding | 0.30 | — |
| aider_polyglot | 0.30 | — |
| swe_bench_verified | 0.40 | 88.9 |
missing signals are dropped and the remaining weights renormalized — never imputed.
Agentic: #4 — score 88.9 ● low — a single signal; treat with caution
| signal | weight | input (0–100) |
|---|---|---|
| swe_bench_verified | 0.60 | 88.9 |
| swe_rebench | 0.40 | — |
missing signals are dropped and the remaining weights renormalized — never imputed.
# trend
methodology changed during this history window; score movement across that boundary is not model movement. See methodology v5.
# benchmarks
| benchmark | score | source | date |
|---|---|---|---|
| AA Intelligence Index | 17.2 | Artificial Analysis | — |
| AA Math Index | 57.3 | Artificial Analysis | — |
| AIME 2024 | 0.7 / 1 | LLM Stats | — |
| SWE-bench Verified | 71.2 | SWE-bench | Oct 2025 |
# where to run
| provider | quant | ctx | $/M in | $/M out | $/M cache | price src | tps | uptime | ✓ |
|---|---|---|---|---|---|---|---|---|---|
| DeepInfra | 256k | $0.50 | $2.00 | — | via litellm | — | — | ||
| BaseTen | — | $0.60 | $2.50 | — | via litellm | — | — | ||
| Novita | fp8 | 256k | $0.60 | $2.50 | — | novita | — | 100.0% | ✓ |
| Groq | 256k | $1.00 | $3.00 | — | via litellm | — | — | ||
| Together | 256k | $1.00 | $3.00 | — | via litellm | — | — |
sorted by blended price ((3·input + output) / 4 per 1M) · ✓ = the provider's own catalog confirms the offer · "via …" prices are what the aggregator routing the offer charges, not the provider's own list price
source aliases
- aa
kimi-k2-0905- openrouter
moonshotai/kimi-k2-0905- swebench
kimi-k2-0905-preview