W&B Inference models & pricing
W&B Inference hosts 43 models (35 with public pricing) covering 1 modalities. Weights & Biases serverless inference for Llama and DeepSeek. Cheapest input starts at $0.0300/M tokens; the most premium goes up to $1.31/M. Use Future AGI's Agent Command Center to route any W&B Inference model with cost-optimized fallback and unified observability.
chat 43
| Model ↕ | Input / 1M ↕ | Output / 1M ↕ | Blended ↕ | Context ↕ | Caps |
|---|---|---|---|---|---|
| gpt-oss 120B | $0.0300/M | $0.170/M | $0.0650/M | 131,000 | reasoning |
| gpt-oss 20B | $0.0300/M | $0.130/M | $0.0550/M | 131,000 | reasoning |
| Granite 4.1 8B Deprecated | $0.0500/M | $0.1000/M | $0.0625/M | 131,000 | |
| Jetbrains Mellum2 12B A2 5B Instruct Deprecated | $0.0500/M | $0.1000/M | $0.0625/M | 131,000 | |
| Openpipe Qwen3 14B Instruct Deprecated | $0.0500/M | $0.220/M | $0.0925/M | 32,800 | |
| Nvidia Nvidia Nemotron 3.5 Lightning 30B A3B | $0.0700/M | $0.200/M | $0.103/M | 262,000 | reasoning · cache |
| Gemma 4 26B A4B IT | $0.1000/M | $0.300/M | $0.150/M | 262,000 | tools · vision · reasoning · cache |
| Gemma 4 31B IT | $0.1000/M | $0.340/M | $0.160/M | 262,000 | vision · reasoning |
| Granite 4.2 8B | $0.1000/M | $0.150/M | $0.112/M | 131,000 | reasoning · cache |
| Qwen3 30B A3B Instruct 2507 Deprecated | $0.1000/M | $0.300/M | $0.150/M | 262,000 | |
| DeepSeek V4 Flash 0731 | $0.130/M | $0.280/M | $0.168/M | 262,000 | reasoning · cache |
| DeepSeek V4 Flash Deprecated | $0.140/M | $0.280/M | $0.175/M | 1,049,000 | reasoning · cache |
| Nvidia Nvidia Nemotron 3.5 Super VL 120B A12B | $0.150/M | $0.750/M | $0.300/M | 1,049,000 | tools · vision · reasoning · cache |
| GLM 5.3 Flash | $0.150/M | $0.500/M | $0.238/M | 1,049,000 | tools · vision · reasoning · cache |
| DeepSeek V4.1 Flash | $0.200/M | $0.650/M | $0.313/M | 1,049,000 | vision · reasoning · cache |
| Llama 3.1 8B Instruct | $0.220/M | $0.220/M | $0.220/M | 131,000 | |
| MiniMax M3 | $0.230/M | $0.960/M | $0.413/M | 262,000 | vision · reasoning · cache |
| Qwen3.5 35B A3B Deprecated | $0.250/M | $1.25/M | $0.500/M | 262,000 | vision · reasoning |
| Qwen3.6 35B A3B | $0.250/M | $1.25/M | $0.500/M | 262,000 | vision · reasoning |
| MiniMax M2.5 | $0.300/M | $1.20/M | $0.525/M | 197,000 | tools · reasoning |
| Qwen3.8 27B | $0.400/M | $3.00/M | $1.05/M | 262,000 | vision · reasoning · cache |
| Nvidia Nvidia Nemotron 3 Ultra 550B A55B | $0.500/M | $2.15/M | $0.913/M | 262,000 | reasoning · cache |
| DeepSeek v3.1 | $0.550/M | $1.65/M | $0.825/M | 161,000 | reasoning |
| Kimi K2.5 | $0.600/M | $3.00/M | $1.20/M | 262,144 | tools · vision · reasoning |
| Kimi K2 Instruct | $0.600/M | $2.50/M | $1.08/M | 128,000 | |
| Qwen3.6 27B Deprecated | $0.600/M | $3.60/M | $1.35/M | 262,000 | vision · reasoning · cache |
| Kimi K2.6 | $0.650/M | $3.41/M | $1.34/M | 262,000 | vision · reasoning · cache |
| Llama 3.3 70B Instruct | $0.710/M | $0.710/M | $0.710/M | 128,000 | |
| Kimi K2.7 Code | $0.710/M | $3.50/M | $1.41/M | 262,000 | vision · reasoning · cache |
| GLM 5.2 | $0.760/M | $2.42/M | $1.18/M | 1,049,000 | reasoning · cache |
| Llama 3.1 70B Instruct Deprecated | $0.800/M | $0.800/M | $0.800/M | 131,000 | |
| Reflection Beam 1.0 501B A23B | $1.00/M | $3.00/M | $1.50/M | 1,049,000 | tools · reasoning · cache |
| DeepSeek V4 Pro Deprecated | $1.15/M | $2.55/M | $1.50/M | 1,049,000 | reasoning · cache |
| GLM 5.3 | $1.20/M | $4.00/M | $1.90/M | 1,049,000 | tools · reasoning · cache |
| DeepSeek V4 Pro 0813 | $1.31/M | $3.96/M | $1.97/M | 1,049,000 | reasoning · cache |
| DeepSeek R1 0528 | — | — | — | 161,000 | |
| DeepSeek v3.0324 | — | — | — | 161,000 | |
| GLM 4.5 | — | — | — | 131,072 | |
| Llama 4 Scout 17B 16e Instruct | — | — | — | 64,000 | |
| Phi 4 mini Instruct | — | — | — | 128,000 | |
| Qwen3 235B A22B Instruct 2507 | — | — | — | 262,144 | |
| Qwen3 235B A22B Thinking 2507 | — | — | — | 262,144 | |
| Qwen3 Coder 480B A35B Instruct | — | — | — | 262,144 |
FAQ
How many W&B Inference models are there?
43 W&B Inference models are listed across 1 modality on this page. 35 have public per-token pricing.
How is W&B Inference pricing verified?
Pricing is aggregated from BerriAI/litellm, models.dev, and OpenRouter and refreshed weekly. Each row shows a per-model "verified" date. If a price is wrong, click the row to open the model page and use the inline "suggest edit" link — submissions go into a public review queue.
Which W&B Inference model is cheapest?
Input pricing on W&B Inference starts at $0.0300 per 1M tokens. Sort the table by price (or use the in-page filter at the top) to find the cheapest model that matches your capability requirements.
Can I route to W&B Inference via an OpenAI-compatible API?
Yes — point your OpenAI client at Future AGI's Agent Command Center, configure a W&B Inference target, and call W&B Inference models with the standard /v1/chat/completions surface. The same gateway can route to other providers as fallback. Free for the first 100K requests/month.