Largest Gemma 4 instruction model for open, self-hosted chat and reasoning
Benchmarks
| Benchmark | Category | Score | Variant | Provenance | Source | Observed |
|---|---|---|---|---|---|---|
| AA Agentic Index | agentic | 4.2 | — | mirror | openrouter | Sep 20, 2026 |
| AA Coding Index | coding | 43.4 | — | mirror | openrouter | Aug 7, 2026 |
| AA Intelligence Index | intelligence | 4.4 | — | mirror | openrouter | Nov 5, 2026 |
Includes index data from Artificial Analysis.
Providers
Live operational stats per hosting endpoint, via OpenRouter.
| Provider | Quant | Context | $/1M in · out | Uptime 24h | Latency | Throughput |
|---|---|---|---|---|---|---|
| Chutes | fp4 | 131K | $0.12 · $0.37 | 99.08% | — | — |
| CoreWeave | fp4 | 262K | $0.1 · $0.34 | 100.00% | — | — |
| Crusoe | bf16 | 262K | $0.14 · $0.4 | 100.00% | — | — |
| DeepInfra | fp8 | 262K | $0.2 · $0.4 | 100.00% | — | — |
| DeepInfra | fp8 | 262K | $0.16 · $0.32 | 100.00% | — | — |
| DeepInfra | fp8 | 262K | $0.3 · $0.6 | 99.71% | — | — |
| DeepInfra | fp4 | 262K | $0.135 · $0.51 | 100.00% | — | — |
| Friendli | — | 262K | $0.14 · $0.4 | 100.00% | — | — |
| KT | — | 262K | $0.1 · $0.34 | 99.65% | — | — |
| Modal | — | 131K | $0.34 · $0.1 | 100.00% | — | — |
| ModelRun | fp4 | 262K | $0.75 · $1 | 100.00% | — | — |
| NextBit | fp4 | 262K | $0.125 · $0.34 | 100.00% | — | — |
| Novita | bf16 | 262K | $0.14 · $0.4 | 97.99% | — | — |
| Parasail | fp8 | 262K | $0.15 · $0.4 | 99.37% | — | — |
| PrimeIntellect | fp8 | 262K | $0.2 · $0.335 | 99.86% | — | — |
| SambaNova | — | 131K | $0.38 · $1.15 | 97.38% | — | — |
| SiliconFlow | fp8 | 262K | $0.75 · $1 | 99.08% | — | — |
| Venice | fp4 | 256K | $0.12 · $0.36 | 100.00% | — | — |