Cheapest LLM API, ranked by price
CoreWeave is the cheapest direct API for gpt-oss-120b: $0.03 input and $0.17 output per 1M tokens. OpenRouter undercuts it at $0.03 / $0.15 by routing to the cheapest host, before its 5.5% credit fee. The same model costs up to $0.75 output elsewhere, so the provider you pick moves your bill by 7×. Prices as of .
Low: OpenRouter $0.15 High: Parasail, Cerebras, Cloudflare Workers AI $0.75
gpt-oss-120b: 21 providers, cheapest first
USD per 1M tokens| # | Provider | Input | Output | Blended 3:1 | 10M in + 2M out | Hardware | Checked |
|---|---|---|---|---|---|---|---|
| 1 | OpenRouterrouter | $0.03 | $0.15 | $0.06 | $0.60 | — | 2026-11-09 |
| 2 | CoreWeave | $0.03 | $0.17 | $0.065 | $0.64 | GPU | 2026-11-09 |
| 3 | DeepInfra | $0.037 | $0.17 | $0.07 | $0.71 | GPU | 2026-11-09 |
| 4 | Hugging Face Inference Providersrouter | $0.037 | $0.17 | $0.07 | $0.71 | — | 2026-11-09 |
| 5 | Crusoe | $0.05 | $0.25 | $0.10 | $1.00 | GPU | 2026-11-09 |
| 6 | Novita AI | $0.05 | $0.25 | $0.10 | $1.00 | GPU | 2026-11-09 |
| 7 | SiliconFlow | $0.05 | $0.45 | $0.15 | $1.40 | GPU | 2026-11-09 |
| 8 | Google Vertex AI | $0.09 | $0.36 | $0.158 | $1.62 | Mixed | 2026-11-09 |
| 9 | Baseten | $0.10 | $0.50 | $0.20 | $2.00 | GPU | 2026-11-09 |
| 10 | Amazon Bedrock | $0.15 | $0.60 | $0.262 | $2.70 | — | 2026-11-09 |
| 11 | Azure AI Foundry | $0.15 | $0.60 | $0.262 | $2.70 | — | 2026-11-09 |
| 12 | Featherless AI | $0.15 | $0.60 | $0.262 | $2.70 | GPU | 2026-11-09 |
| 13 | Fireworks AI | $0.15 | $0.60 | $0.262 | $2.70 | GPU | 2026-11-09 |
| 14 | Groq | $0.15 | $0.60 | $0.262 | $2.70 | Custom silicon | 2026-11-09 |
| 15 | Nebius Token Factory | $0.15 | $0.60 | $0.262 | $2.70 | GPU | 2026-11-09 |
| 16 | Together AI | $0.15 | $0.60 | $0.262 | $2.70 | GPU | 2026-11-09 |
| 17 | Parasail | $0.10 | $0.75 | $0.263 | $2.50 | GPU | 2026-11-09 |
| 18 | SambaNova | $0.22 | $0.59 | $0.313 | $3.38 | Custom silicon | 2026-11-09 |
| 19 | Replicate | $0.18 | $0.72 | $0.315 | $3.24 | GPU | 2026-11-09 |
| 20 | Cerebras | $0.35 | $0.75 | $0.45 | $5.00 | Custom silicon | 2026-11-09 |
| 21 | Cloudflare Workers AI | $0.35 | $0.75 | $0.45 | $5.00 | GPU | 2026-11-09 |
Input and output link the provider’s price page. Blended is (3 × input + output) ÷ 4, the same 3:1 input-to-output mix Artificial Analysis uses for its blended price. The workload column is 10M input plus 2M output tokens, priced from the listed rates. Standard serverless tier only; batch, priority and cached-input rates are on each provider’s page.
No published per-token gpt-oss-120b price: Alibaba Cloud Model Studio, FriendliAI, Hyperbolic, Mistral AI, Modal, Nscale, Scaleway Generative APIs.
Llama 3.3 70B: 11 providers still list it
USD per 1M tokens| # | Provider | Input | Output | Blended 3:1 | Checked |
|---|---|---|---|---|---|
| 1 | DeepInfra | $0.10 | $0.32 | $0.155 | 2026-11-09 |
| 2 | OpenRouter | $0.10 | $0.32 | $0.155 | 2026-11-09 |
| 3 | Hugging Face Inference Providers | $0.135 | $0.40 | $0.201 | 2026-11-09 |
| 4 | Novita AI | $0.135 | $0.40 | $0.201 | 2026-11-09 |
| 5 | Parasail | $0.22 | $0.50 | $0.29 | 2026-11-09 |
| 6 | Azure AI Foundry | $0.71 | $0.71 | $0.71 | 2026-11-09 |
| 7 | CoreWeave | $0.71 | $0.71 | $0.71 | 2026-11-09 |
| 8 | Amazon Bedrock | $0.72 | $0.72 | $0.72 | 2026-11-09 |
| 9 | Google Vertex AI | $0.72 | $0.72 | $0.72 | 2026-11-09 |
| 10 | SambaNova | $0.60 | $1.20 | $0.75 | 2026-11-09 |
| 11 | Cloudflare Workers AI | $0.293 | $2.253 | $0.783 | 2026-11-09 |
Together, Cerebras, Baseten, Crusoe and SiliconFlow have taken Llama 3.3 70B off serverless, and Groq moved it to contact-sales. That is why gpt-oss-120b is the reference.
For steady high volume: dedicated H100s
USD per GPU-hour, on demand| # | Provider | H100 / hour | 30 days, 24/7 | Checked |
|---|---|---|---|---|
| 1 | DeepInfra | $2.90 | $2,088 | 2026-11-09 |
| 2 | Modal | $3.95 | $2,844 | 2026-11-09 |
| 3 | FriendliAI | $5.00 | $3,600 | 2026-11-09 |
| 4 | Replicate | $5.49 | $3,953 | 2026-11-09 |
| 5 | Together AI | $5.49 | $3,953 | 2026-11-09 |
| 6 | Crusoe | $5.50 | $3,960 | 2026-11-09 |
| 7 | Baseten | $6.50 | $4,680 | 2026-11-09 |
| 8 | Fireworks AI | $8.00 | $5,760 | 2026-11-09 |
A dedicated GPU bills whether it is busy or idle. It beats per-token pricing once your steady traffic keeps it near full; below that, serverless is cheaper.
Price is one column
Full provider tableThe main table puts these prices next to hardware, vendor-stated speed, OpenAI compatibility, SOC 2 and zero data retention. Filter to the providers that pass your security review, then sort by price.