The board shows this model's cheapest measured offer ($0.068/1M); the smaller figure under it is the mean of the 26 offers listed below ($0.167/1M). Add those up and divide — it matches. 24h token volume (55B) is sourced from OpenRouter public rankings — today's only public source of absolute per-model token counts (Vercel AI Gateway publishes share-of-traffic percentages); as more sources come online, the per-router breakdown will sum to this total here.
| Provider | Router / Source | $ in /1M | $ out /1M | TTFT | TPS | Uptime | Evidence |
|---|---|---|---|---|---|---|---|
| Darkbloom | OpenRouter ↗ | $0.040 | $0.150 | 660ms | 50 | 100% | API-measured |
| Makora | OpenRouter ↗ | $0.080 | $0.100 | 385ms | 106 | 100% | API-measured |
| requesty | Requesty ↗ | $0.056 | $0.272 | — | — | — | API-measured |
| PrimeIntellect (bf16) | OpenRouter ↗ | $0.075 | $0.225 | 615ms | 52 | 100% | API-measured |
| AkashML (fp4) | OpenRouter ↗ | $0.070 | $0.240 | 405ms | 121 | 100% | API-measured |
| DekaLLM (bf16) | OpenRouter ↗ | $0.060 | $0.330 | 583ms | 64 | 100% | API-measured |
| deepinfra | Pricebook (LiteLLM) ↗ | $0.070 | $0.340 | — | — | — | Official list |
| deepinfra | HuggingFace Router ↗ | $0.070 | $0.340 | 190ms | 91 | — | API-measured |
| DeepInfra (fp8) | OpenRouter ↗ | $0.070 | $0.340 | 468ms | 53 | 100% | API-measured |
| NextBit (bf16) | OpenRouter ↗ | $0.090 | $0.300 | 292ms | 76 | 100% | API-measured |
| Io Net (bf16) | OpenRouter ↗ | $0.090 | $0.310 | 485ms | 104 | 100% | API-measured |
| cloudflare | Pricebook (LiteLLM) ↗ | $0.100 | $0.300 | — | — | — | Official list |
| wandb | Pricebook (LiteLLM) ↗ | $0.100 | $0.300 | — | — | — | Official list |
| CoreWeave (bf16) | OpenRouter ↗ | $0.100 | $0.300 | 218ms | 65 | 100% | API-measured |
| Cloudflare | OpenRouter ↗ | $0.100 | $0.300 | 578ms | 42 | 100% | API-measured |
| novita | Pricebook (LiteLLM) ↗ | $0.130 | $0.400 | — | — | — | Official list |
| novita | HuggingFace Router ↗ | $0.130 | $0.400 | 703ms | 103 | — | API-measured |
| novita-ai | Novita AI ↗ | $0.130 | $0.400 | — | — | — | API-measured |
| Venice (bf16) | OpenRouter ↗ | $0.130 | $0.400 | 865ms | 68 | 100% | API-measured |
| Parasail (bf16) | OpenRouter ↗ | $0.130 | $0.400 | 462ms | 92 | 100% | API-measured |
| Novita (bf16) | OpenRouter ↗ | $0.130 | $0.400 | 976ms | 124 | 100% | API-measured |
| aihubmix | Pricebook (LiteLLM) ↗ | $0.140 | $0.400 | — | — | — | Official list |
| SiliconFlow (fp8) | OpenRouter ↗ | $0.140 | $0.400 | 1102ms | 67 | 100% | API-measured |
| vertex_ai | Pricebook (LiteLLM) ↗ | $0.150 | $0.600 | — | — | — | Official list |
| Google (global) | OpenRouter ↗ | $0.150 | $0.600 | 1769ms | 35 | 100% | API-measured |
| scaleway | Pricebook (LiteLLM) ↗ | $0.250 | $0.500 | — | — | — | Official list |
Volume trend
Last 29 days: 53B → 55B (+4%); range 0–63B tokens/day. Source: OpenRouter public rankings; refreshed hourly.
Data & method
Cheapest ranked by blended = (3×in + out) ÷ 4; performance columns show only when a source provides them, never estimated. “Official list” = LiteLLM open pricebook (MIT); “API-measured” = provider/aggregator public APIs. Machine-readable data: /api/v1/models/gemma-4-26b-a4b-it.json. Free reuse requires attribution to tkx.org.
About
Gemma 4 26B A4B is a large language model served by multiple inference providers on AI model marketplaces. It is used for natural language processing tasks, leveraging its capabilities in understanding and generating human-like text.
AI-generated summary from public information — is this yours?
FAQ
How much does Gemma 4 26B A4B (free) cost?
Cheapest measured offer right now: $0.04/1M input, $0.15/1M output via Darkbloom — across 19 tracked providers, refreshed hourly.
Who serves Gemma 4 26B A4B (free)?
19 providers currently serve it. The table above lists each with input/output price, TTFT, TPS, uptime and an evidence tier.
How heavily is Gemma 4 26B A4B (free) used?
Gemma 4 26B A4B (free) routed 55B tokens in the last 24h. The daily bars above show routed volume over recent days. Source: OpenRouter public rankings — one of the few routers publishing per-model usage; TKX refreshes hourly.