CoreWeave Forge (formerly W&B Inference) API pricing
Updated
CoreWeave Forge (formerly W&B Inference) lists standard prices for 20 models. The cheapest is GPT OSS 20B at USD 0.03 input and USD 0.13 output per 1M tokens; the dearest is DeepSeek V4 Pro 0813 at USD 1.31 / USD 3.96. It publishes cached-input prices for 15 of them, and lists no per-model batch, flex or priority prices on the pages we read.
- Models with a standard price
- 20
- Median input / output
- USD 0.175 / USD 0.68
- Cached-input prices
- 15 of 20
- Cheapest host for
- 7 of 20 shared models
Which models is CoreWeave Forge (formerly W&B Inference) the cheapest provider for?
Of the 20 models that CoreWeave Forge (formerly W&B Inference) and at least one other provider both price at the standard tier, CoreWeave Forge (formerly W&B Inference) has the lowest price for 7, ranked on a blend of 3 input tokens to 1 output token. It is the dearest for 1.
| Model | CoreWeave Forge (formerly W&B Inference) input / output | Next cheapest | Saving on 3:1 blend | Providers |
|---|---|---|---|---|
| OpenAI GPT OSS 120B | USD 0.03 / USD 0.17 | DeepInfra USD 0.037 / USD 0.17 | 7% | 21 |
| Kimi K2.7 Code | USD 0.60 / USD 3 | Parasail USD 0.75 / USD 3.50 | 17% | 12 |
| Gemma 4 31B | USD 0.10 / USD 0.34 | Featherless USD 0.12 / USD 0.36 | 11% | 12 |
| MiniMax M3 | USD 0.23 / USD 0.96 | GMI Cloud USD 0.24 / USD 0.96 | 2% | 10 |
| Qwen3.6 35B A3B | USD 0.10 / USD 0.89 | AkashML USD 0.10 / USD 0.90 | 1% | 10 |
| NVIDIA Nemotron 3 Ultra | USD 0.50 / USD 2.15 | DeepInfra USD 0.50 / USD 2.20 | 1% | 9 |
| IBM Granite 4.2 8B | USD 0.06 / USD 0.15 | DeepInfra USD 0.06 / USD 0.25 | 23% | 2 |
How does CoreWeave Forge (formerly W&B Inference) price the most widely hosted models?
The 6 models CoreWeave Forge (formerly W&B Inference) sells that the most other providers also sell, with CoreWeave Forge (formerly W&B Inference)'s rank among them on the 3:1 blend.
| Model | CoreWeave Forge (formerly W&B Inference) | Rank | Cheapest |
|---|---|---|---|
| OpenAI GPT OSS 120B | USD 0.03 / USD 0.17 | 1 of 21 | CoreWeave Forge (formerly W&B Inference) (this page) |
| GLM 5.3 Flash | USD 0.12 / USD 0.38 | 2 of 20 | GMI Cloud USD 0.075 / USD 0.25 |
| GLM 5.2 | USD 0.62 / USD 2.42 | 2 of 19 | DeepInfra USD 0.5625 / USD 1.80 |
| DeepSeek V4.1 Flash | USD 0.13 / USD 0.65 | 2 of 16 | DeepInfra USD 0.14 / USD 0.42 |
| DeepSeek V4 Pro 0813 | USD 1.31 / USD 3.96 | 4 of 15 | Chutes USD 0.85 / USD 2.65 |
| DeepSeek V4 Flash 0731 | USD 0.13 / USD 0.27 | 5 of 15 | DeepInfra USD 0.06 / USD 0.18 |
Which model makers does CoreWeave Forge (formerly W&B Inference) sell?
CoreWeave Forge (formerly W&B Inference) prices models from 10 makers. DeepSeek leads with 4 models. Standard prices, USD per 1M input / output tokens.
| Maker | Models | Cheapest model | Median input / output |
|---|---|---|---|
| DeepSeek | 4 | DeepSeek V4 Flash 0731 USD 0.13 / USD 0.27 | USD 0.315 / USD 1.075 |
| 2 | Gemma 4 26B A4B USD 0.10 / USD 0.30 | USD 0.10 / USD 0.32 | |
| Meta | 2 | Llama 3.1 8B Instruct USD 0.22 / USD 0.22 | USD 0.465 / USD 0.465 |
| Moonshot AI | 2 | Kimi K2.7 Code USD 0.60 / USD 3 | USD 0.625 / USD 3.205 |
| NVIDIA | 2 | NVIDIA Nemotron 3.5 Lightning 30B A3B USD 0.07 / USD 0.20 | USD 0.285 / USD 1.175 |
| OpenAI | 2 | GPT OSS 20B USD 0.03 / USD 0.13 | USD 0.03 / USD 0.15 |
| Qwen | 2 | Qwen3.6 35B A3B USD 0.10 / USD 0.89 | USD 0.25 / USD 1.945 |
| Z.ai | 2 | GLM 5.3 Flash USD 0.12 / USD 0.38 | USD 0.37 / USD 1.40 |
| IBM | 1 | IBM Granite 4.2 8B USD 0.06 / USD 0.15 | — |
| MiniMax | 1 | MiniMax M3 USD 0.23 / USD 0.96 | — |
CoreWeave Forge (formerly W&B Inference) price list
USD per 1M tokens: 20 prices for 20 models, one row per model with any other service tiers stacked under the standard price. Sources checked 2026-11-15 unless a row says otherwise. "vs maker" compares a standard price with the model maker's own API on the 3:1 blend; a minus means CoreWeave Forge (formerly W&B Inference) is cheaper. Of 9 hosted model prices with a maker price to compare, 8 are below the maker's and 0 above.
| Model | Tier | Input | Cached | Output | Context | vs maker | Source |
|---|---|---|---|---|---|---|---|
| DeepSeek V3.1 | standard | USD 0.50 | USD 0.20 | USD 1.50 | 161,000 | — | source |
| DeepSeek V4 Flash 0731 | standard | USD 0.13 | USD 0.05 | USD 0.27 | 262,144 | — | source |
| DeepSeek V4 Pro 0813 | standard | USD 1.31 | USD 0.044 | USD 3.96 | 1,048,576 | same | source |
| DeepSeek V4.1 Flash | standard | USD 0.13 | USD 0.002 | USD 0.65 | 1,048,576 | −50% | source |
| Gemma 4 26B A4B | standard | USD 0.10 | USD 0.05 | USD 0.30 | 262,144 | — | source |
| Gemma 4 31B | standard | USD 0.10 | — | USD 0.34 | 262,144 | — | source |
| GLM 5.2 | standard | USD 0.62 | USD 0.14 | USD 2.42 | 1,048,576 | −50% | source |
| GLM 5.3 Flash | standard | USD 0.12 | USD 0.02 | USD 0.38 | 1,048,576 | −22% | source |
| GPT OSS 20B | standard | USD 0.03 | — | USD 0.13 | 131,072 | — | source |
| IBM Granite 4.2 8B | standard | USD 0.06 | USD 0.015 | USD 0.15 | 131,072 | — | source |
| Kimi K2.6 | standard | USD 0.65 | USD 0.15 | USD 3.41 | 262,144 | −22% | source |
| Kimi K2.7 Code | standard | USD 0.60 | USD 0.12 | USD 3 | 262,144 | −30% | source |
| Llama 3.1 8B Instruct | standard | USD 0.22 | — | USD 0.22 | 131,072 | — | source |
| Llama 3.3 70B | standard | USD 0.71 | — | USD 0.71 | 128,000 | — | source |
| MiniMax M3 | standard | USD 0.23 | USD 0.05 | USD 0.96 | 262,144 | −21% | source |
| NVIDIA Nemotron 3 Ultra | standard | USD 0.50 | USD 0.10 | USD 2.15 | 262,144 | — | source |
| NVIDIA Nemotron 3.5 Lightning 30B A3B | standard | USD 0.07 | USD 0.04 | USD 0.20 | 262,144 | — | source |
| OpenAI GPT OSS 120B | standard | USD 0.03 | — | USD 0.17 | 131,072 | — | source |
| Qwen3.6 35B A3B | standard | USD 0.10 | USD 0.035 | USD 0.89 | 262,144 | −65% | source |
| Qwen3.8 27B | standard | USD 0.40 | USD 0.15 | USD 3 | 262,144 | −7% | source |
- Cached "—" means CoreWeave Forge (formerly W&B Inference)'s page shows no cache price for that row; budget cached tokens at the input price.
Report a wrong price or number (email to info@broadnet.ai, with this page's link filled in).
What service tiers does CoreWeave Forge (formerly W&B Inference) publish?
Standard prices only: CoreWeave Forge (formerly W&B Inference) publishes no per-model batch, flex, priority or fast price on the pages we read.
Cached input: 15 of 20 standard prices include a cache-read price, a median 23% of the input price. What cached input means.
What we recorded about CoreWeave Forge (formerly W&B Inference)'s pricing page: wandb.ai/site/pricing/inference now redirects to CoreWeave Forge pricing. Prices per 1M tokens from the "CoreWeave Forge hosted models" table; a "-" cache-hit cell is recorded as no cached price. Model IDs and context windows from docs.coreweave.com/products/inference/serverless/models (context "1049k" read as 1,048,576; "128k" as 128,000; "161k" as 161,000). Reflection Beam 1.0 and DeepSeek V4.1-Flash-1017 are listed as available models but have no price on the pricing page, so they are left out. ARIA agent rates (OpenAI models plus an agent fee) excluded: not a model API price. Each Forge plan includes a monthly inference allowance used before pay-as-you-go billing.
Is CoreWeave Forge (formerly W&B Inference) cheaper than the model makers' own APIs?
Mostly yes. Of 9 standard prices for models whose maker sells its own API, 8 are below the maker's price, 0 above and 1 the same, on the 3:1 blend. The median difference is 22% below the maker. The biggest saving is Qwen3.6 35B A3B, 65% below Alibaba Cloud.
BroadNet Inference and CoreWeave Forge (formerly W&B Inference)
BroadRouter is BroadNet's own site, so BroadNet is never ranked on this page. BroadNet Inference also serves 3 models that CoreWeave Forge (formerly W&B Inference) prices: BroadNet Gemma 4 VL 26B at USD 0.10 / USD 0.22 against CoreWeave Forge (formerly W&B Inference)'s USD 0.10 / USD 0.30 (BroadNet is cheaper on the 3:1 blend); DeepSeek V4.1 Flash at USD 0.175 / USD 0.52 against CoreWeave Forge (formerly W&B Inference)'s USD 0.13 / USD 0.65 (CoreWeave Forge (formerly W&B Inference) is cheaper on the 3:1 blend); Qwen3.8 27B at USD 0.175 / USD 2.17 against CoreWeave Forge (formerly W&B Inference)'s USD 0.40 / USD 3 (BroadNet is cheaper on the 3:1 blend). BroadNet prices are from its rate card, synced 2026-11-14. BroadNet Inference pricing.
More on CoreWeave Forge (formerly W&B Inference)
- CoreWeave Forge (formerly W&B Inference)'s own pricing page (the source; checked 2026-11-15)
- All 41 providers compared · Cheapest LLM APIs right now · Download every price (CSV, JSON)
Questions
How much does the CoreWeave Forge (formerly W&B Inference) API cost per token?
CoreWeave Forge (formerly W&B Inference)'s standard prices run from USD 0.03 input and USD 0.13 output per 1M tokens (GPT OSS 20B) to USD 1.31 / USD 3.96 (DeepSeek V4 Pro 0813). The median across 20 standard prices is USD 0.175 input and USD 0.68 output per 1M, which is USD 0.000000175 per input token. Input, cached input and output are billed separately.
What is the cheapest model on CoreWeave Forge (formerly W&B Inference)?
GPT OSS 20B, at USD 0.03 input and USD 0.13 output per 1M tokens (USD 0.055 on a 3:1 input-to-output blend). The cheapest from other makers: DeepSeek V4 Flash 0731 (DeepSeek) at USD 0.13 / USD 0.27; Gemma 4 26B A4B (Google) at USD 0.10 / USD 0.30; Llama 3.1 8B Instruct (Meta) at USD 0.22 / USD 0.22.
Is CoreWeave Forge (formerly W&B Inference) cheaper than other providers?
It depends on the model. Of 20 models that other providers also price, CoreWeave Forge (formerly W&B Inference) is the cheapest standard price for 7, for example OpenAI GPT OSS 120B, Kimi K2.7 Code and Gemma 4 31B; another provider is cheaper for 13; and it is the dearest for 1. Its median rank is 2 of a median 12 providers, on a 3:1 input-to-output blend.
Does CoreWeave Forge (formerly W&B Inference) have batch, cached or priority pricing?
Cached input: yes, for 15 of 20 standard prices, at a median 23% of the input price. No batch, flex or priority prices are listed per model on the pages we read.
How much does a chatbot cost per month on CoreWeave Forge (formerly W&B Inference)?
For 2,000 requests a day, 1,500 input tokens (60% cached), 300 output, over 30 days: USD 5.04 on GPT OSS 20B, USD 21.51 on Qwen3.6 35B A3B (the middle of CoreWeave Forge (formerly W&B Inference)'s list) and USD 120.82 on DeepSeek V4 Pro 0813. Try your own numbers.