Fireworks AI: serverless and on-demand hosting now in our directory

We've added Fireworks AI to LLM Hosting Pro. Fireworks is one of the most popular platforms for running open-weight models, and you can now compare its hosting plans, prices and supported models alongside the other LLM hosts we track.

Who's behind Fireworks AI

Fireworks AI was founded in 2022 and is based in Redwood City, California. Its CEO, Lin Qiao, previously led PyTorch at Meta. She started the company with six co-founders from Meta's PyTorch and infrastructure teams and Google's Vertex AI: Benny Chen, Chenyu Zhao, Dmytro Dzhulgakov, Dmytro Ivchenko, James Reed and Pawel Garbacki (source). That PyTorch background shows in the product: Fireworks runs its own inference engine, built for fast, low-cost serving of open models.

The hosting plans we've listed

We've added two Fireworks plans. Prices are from Fireworks' pricing page and change often, so check there for current rates.

• Fireworks Serverless: an OpenAI-compatible API billed per token, from about US$0.05 to US$22.50 per million tokens. Most open models follow simple size-based prices, for example US$0.20 per million tokens for 4B to 16B models and US$0.90 above 16B, and batch jobs cost half the serverless price.
• Fireworks On-Demand Deployments: your model on dedicated NVIDIA GPUs, billed per GPU-second, from US$8 an hour for an H100 or H200 to US$20 an hour for a GB300.

Fireworks also offers fine-tuning (LoRA and full), and fine-tuned models are served at the same price as their base model.

Models

Fireworks runs 17 of the models in our directory, across six model families: Llama, Mistral, DeepSeek, Qwen, Gemma and gpt-oss. These include Llama 3.3 70B, DeepSeek-R1 and V3.1, Qwen3 235B-A22B, Gemma 3 27B and gpt-oss-120b. Its full model catalogue also covers newer families we don't list yet, such as Kimi, GLM and MiniMax.

Who uses Fireworks

Fireworks' customer list includes some well-known AI products:

• Cursor uses Fireworks' speculative decoding for its Fast Apply code-editing feature.
• Notion fine-tuned models on Fireworks and cut latency from about 2 seconds to 350 milliseconds.
• Vercel uses a fine-tuned model on Fireworks as part of its v0 app builder.
• Cresta runs its Knowledge Assist agent guidance on Fireworks, and reports costs up to 100 times lower than GPT-4.
• Upwork, Sourcegraph, Quora and Genspark also run models on Fireworks.

Compare Fireworks with other providers on our LLM hosting plans page, or see the full Fireworks AI host profile.