K2 Think V2 k2-think-v2
IFM · ifm
#65 overall
- params
- 72.6B
- arch
- dense
- license
- apache-2.0 open weights
- released
- Jan 2026
- downloads/30d
- 784
# architecture
- layers
- 80
- d_model
- 8192
- heads
- 64q / 8kv
- head dim
- 128
- vocab
- 250112
- family
- llama
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: #65 — score 38.3 ● low — a single signal; treat with caution
| signal | weight | input (0–100) |
|---|---|---|
| aa_intelligence | 0.70 | 40.2 |
| bench_composite | 0.30 | 33.7 |
benchmark panel evidence:
| complete panel | score (0–100) |
|---|---|
| aa-coding-v1 | 33.7 |
Coding: provisional — score 23.5 ● 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 | 23.5 |
| aider_polyglot | 0.30 | — |
| swe_bench_verified | 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 v6.
# benchmarks
| benchmark | score | source | date |
|---|---|---|---|
| AA Coding Index | 21.0 | Artificial Analysis | — |
| AA Intelligence Index | 11.5 | Artificial Analysis | — |
# where to run
no known API providers — download the weights from Hugging Face and run them yourself.
source aliases
- aa
k2-think-v2