Fireworks AI
Serverless inference, GPU deployments and fine-tuning for open models
Overview
Fireworks AI is an inference and training platform for open-weight models such as Kimi, DeepSeek, GLM, Qwen and gpt-oss, served per token through an OpenAI-compatible API or on dedicated GPUs billed per second. It also offers managed fine-tuning, a training API, batch inference and enterprise options such as reserved throughput and data residency. Billing uses prepaid credits.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 14 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- Kimi K3 $3.00 / $0.30 / $15.00
- DeepSeek V4.1 Flash $0.30 / $0.006 / $1.20
- GLM 5.3 $1.40 / $0.26 / $4.40
- gpt-oss-120B $0.15 / $0.015 / $0.60
- Unlisted models by size, $0.10 to $0.90 per 1M tokens
- Batch inference at 50% of serverless
- Kimi K3 Priority $3.75 / $0.375 / $18.75
- Kimi K3 Fast $4.50 / $0.45 / $22.50
- GLM 5.3 Fast $2.10 / $0.39 / $6.60
- Up to 150M parameters $0.008
- 150M to 350M parameters $0.016
- Qwen3 8B $0.10
- LoRA SFT $0.50 (up to 16B) to $10 (over 300B)
- LoRA DPO and full-parameter SFT $1.00 to $20
- Full-parameter DPO $2.00 to $40
- Fine-tuned models served at base-model prices
- H100 80 GB or H200 141 GB $8/hour
- B200 $13/hour
- B300 $15/hour, GB300 $20/hour
- Region-restricted deployments at 1.5x
- Faster speeds and lower costs
- Higher rate limits
Use cases
What you can produce with Fireworks AI
- Open models such as Kimi K3, DeepSeek V4.1 and gpt-oss called through an OpenAI-compatible API
- JSON-schema or grammar-constrained structured outputs and tool calling for agent pipelines
- An open model fine-tuned on your data with LoRA or full-parameter SFT and DPO, then served at base-model prices
- Dedicated H100, H200 or Blackwell GPU deployments billed per second
- Large asynchronous jobs run through the Batch API at half the serverless price
- Video and audio inputs processed by multimodal models
- Spend tracked and capped with prepaid credits and a monthly spend limit
Fireworks is one of the strongest ways to run open models in production without managing GPUs: a large catalogue, clear per-model token prices, Priority and Fast serving paths, batch at half price, and fine-tuning priced from $0.50 per million training tokens. Customer stories from Cursor, Notion and Cresta, plus SOC 2 Type II, HIPAA and ISO certifications, back its enterprise pitch.
It is a developer tool. You choose and evaluate models yourself, there is no free plan, and US-only or region-restricted serving costs 50% more, which matters for ASEAN teams with residency needs. Excellent for teams optimising cost and latency on open models.
What we found
Fireworks AI runs open-weight models for developers and enterprises: serverless inference priced per token, dedicated GPU deployments billed per second, and managed and API-driven fine-tuning, behind an API compatible with OpenAI's. Its seven co-founders, led by chief executive Lin Qiao, previously head of PyTorch at Meta, came mostly from Meta's PyTorch and infrastructure teams. The model library lists 400+ models; the headline set in October 2026 includes Kimi K3, DeepSeek V4.1 Flash, GLM 5.3, Qwen 3.8 Max, MiniMax M3, OpenAI's gpt-oss-120B and Fireworks' own Ember-1, introduced in September.
The company has grown quickly. After a $250 million Series C at a $4 billion valuation in October 2025, it announced a $1.505 billion Series D in July 2026 at a $17.5 billion valuation, led by Atreides Management, Index Ventures and TCV, with Lightspeed and NVIDIA taking part. Fireworks said then that its annualised revenue had passed $1 billion and that it served more than 40 trillion tokens a day; those are company figures relayed by trade press.
The prices are specific. On the Standard path Kimi K3 costs $3.00 per million input tokens and $15.00 per million output, DeepSeek V4.1 Flash $0.30 and $1.20, GLM 5.3 $1.40 and $4.40, and gpt-oss-120B $0.15 and $0.60. Priority serving costs 25% more, US-only routing 50% more, and batch jobs half. H100 and H200 GPUs rent at $8 an hour, B200 at $13 and GB300 at $20. Billing is prepaid credits with an optional monthly spend limit; earlier versions of this page described postpaid billing and a cap that did not stop requests.
Customer stories on the site include Cursor, Cresta, Upwork and Notion, which is quoted cutting latency from about two seconds to 350 milliseconds. Fireworks reports SOC 2 Type II, HIPAA support and three ISO certifications. There is no free plan.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including Fireworks AI's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Fireworks AI unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Fireworks AI directly →
Spotted something out of date? Suggest an update →
Alternatives to Fireworks AI
More in Agents & Automation
