Gemma 4 31B gemma-4-31b-it
google · gemma flagship
#49 overall
- params
- 32.7B
- arch
- dense
- context
- 256k
- license
- apache-2.0 open weights
- released
- Mar 2026
- reasoning
- yes
- downloads/30d
- 8.4M
# architecture
- layers
- 60
- d_model
- 5376
- heads
- 32q / 16kv
- head dim
- 256
- vocab
- 262144
- family
- gemma4_text
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: #49 — score 54.0 ● high — all signals present
| signal | weight | input (0–100) |
|---|---|---|
| aa_intelligence | 0.70 | 52.0 |
| bench_composite | 0.30 | 58.4 |
benchmark panel evidence:
| complete panel | score (0–100) |
|---|---|
| aa-coding-v1 | 63.1 |
| aime-2026-v1 | 53.8 |
Coding: provisional — score 52.9 ● 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 | 52.9 |
| 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 | 43.4 | Artificial Analysis | — |
| AA Intelligence Index | 14.7 | Artificial Analysis | — |
| AIME 2026 | 0.9 / 1 | LLM Stats | — |
| Big-bench-extra-hard | 0.7 / 1 | LLM Stats | — |
# where to run
| provider | quant | ctx | $/M in | $/M out | $/M cache | price src | tps | uptime | ✓ |
|---|---|---|---|---|---|---|---|---|---|
| Featherless | — | — | — | — | via hfrouter | — | — | ||
| Wandb | 262k | $0.10 | $0.34 | — | via litellm | — | — | ||
| CoreWeave | fp4 | 256k | $0.10 | $0.34 | $0.10 | via openrouter | — | 100.0% | |
| Venice | fp4 | 250k | $0.12 | $0.36 | — | venice | — | 100.0% | ✓ |
| Chutes | fp4 | 128k | $0.12 | $0.37 | $0.01 | via openrouter | — | 93.1% | |
| Aihubmix | 256k | $0.14 | $0.40 | — | via litellm | — | — | ||
| Amazon Bedrock | — | $0.14 | $0.40 | — | bedrock | — | — | ✓ | |
| Crusoe | bf16 | 256k | $0.14 | $0.40 | $0.14 | via openrouter | — | 100.0% | |
| Friendli | unknown | 256k | $0.14 | $0.40 | — | via openrouter | — | 100.0% | |
| Friendliai | 256k | $0.14 | $0.40 | — | via litellm | — | — | ||
| Novita | bf16 | 256k | $0.14 | $0.40 | — | novita | — | 99.2% | ✓ |
| Libertai | 256k | $0.15 | $0.40 | — | via litellm | — | — | ||
| Parasail | fp8 | 256k | $0.15 | $0.40 | $0.06 | via openrouter | — | 99.9% | |
| Phala | 256k | $0.15 | $0.46 | — | phala | — | — | ✓ | |
| Tensormesh | 32k | $0.14 | $0.56 | — | via litellm | — | — | ||
| DeepInfra | fp4 | 256k | $0.20 | $0.40 | — | deepinfra | — | 100.0% | ✓ |
| DeepInfra | fp8 | 256k | $0.20 | $0.40 | — | deepinfra | — | 99.6% | ✓ |
| DeepInfra | fp8 | 256k | $0.20 | $0.40 | — | deepinfra | — | 99.4% | ✓ |
| DeepInfra | fp8 | 256k | $0.20 | $0.40 | — | deepinfra | — | 98.3% | ✓ |
| DigitalOcean | 250k | $0.18 | $0.50 | $0.04 | via modelsdev | — | — | ||
| Sail | 250k | $0.40 | $0.60 | — | via litellm | — | — | ||
| Together | 256k | $0.39 | $0.97 | — | via litellm | — | — | ||
| Io Net | unknown | 256k | $0.36 | $1.09 | $0.18 | via openrouter | — | 97.1% | |
| SambaNova | unknown | 128k | $0.38 | $1.15 | — | sambanova | — | 98.2% | ✓ |
| ModelRun | fp4 | 256k | $0.75 | $1.00 | $0.20 | via openrouter | — | 100.0% | |
| SiliconFlow | fp8 | 256k | $0.75 | $1.00 | $0.25 | openrouter | — | 99.3% | ✓ |
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
gemma-4-31b- bedrock
google.gemma-4-31b- epoch
Gemma 4 31B- openrouter
google/gemma-4-31b-it