The home of open models, datasets and Spaces.
Fireworks AI
Fast inference and fine-tuning for open models.
SharkScore
#39 in Model APIs & inference
Why it scores 64.1
Fireworks AI has a SharkScore of 64.1/100, ranked #39 of 273 in Model APIs & inference (updated 2026-11-19).
Each part compares Fireworks AI with other listings in Model APIs & inference. Evidence covers 87% of the score's weight.
- Adoption
- 37 · 30%
- User reviews
- no data · 25%
- Maintenance
- 100 · 20%
- Trust & transparency
- 80 · 15%
- Momentum
- no data · 10%
Paying never changes this score. How ranking works · method v2026-11-18.2
RankShark review of Fireworks AI
Automated evaluation by RankShark · 18 Nov 2026 · method v1.0 · How we review
We were not paid for this review.
As of 2026-11-18, Fireworks AI offers model inference and training APIs with pay-per-token serverless pricing; it no longer accepts or stores new training datasets. [E2, E4]
Fireworks AI is a platform for running open models, fine-tuning models, and deploying them through serverless or dedicated GPU capacity. Its offerings also include Nexus, a paid product for developers and teams. [E2, E3]
Best for
- Teams building applications that need to call or compare open models through an inference API. · home page, docs
- ML engineers and researchers who want serverless or dedicated options for model training. · pricing page
- Enterprise customers seeking inference in their own cloud account or data center for data residency needs. · pricing page
Not for
- Teams that need Fireworks to create and store new training datasets. · changelog
Key features
- Serverless inference. The vendor says serverless inference is billed per token, with no GPUs to manage. · pricing page, docs
- Reserved throughput. The vendor says reserved throughput provides guaranteed throughput on serverless and on-demand models, backed by an SLA. · pricing page
- Dedicated and sovereign deployments. The vendor says customers can deploy on dedicated GPUs or run Fireworks-managed inference inside their cloud account or data center. · pricing page, docs
- Model training. The vendor says its Training API supports custom training loops, and its docs describe supervised and reinforcement learning fine-tuning. · pricing page, docs
- Model catalog. The vendor says the catalog includes more than 100 text, vision, audio, image, and embedding models. · docs
- Inference modes. The vendor lists Standard, Priority, Fast, Ultra-fast, Batch, and Spot serverless modes; Ultra-fast is in preview. · pricing page, changelog
Pros
- Offers both token-priced serverless inference and reserved or dedicated capacity. · pricing page
- Lists multiple inference modes, including Batch and Spot options with stated discounts. · pricing page
- Provides serverless and dedicated training options with per-model or per-GPU pricing. · pricing page
- The vendor says it supports data residency across 30+ regions. · home page
Cons
- The changelog says dataset creation, upload, and download APIs were retired, and Fireworks no longer stores new datasets. · changelog
- Fireworks Sovereign is listed as a private preview for Enterprise customers. · pricing page
Pricing
| Plan | Price | Limits |
|---|---|---|
| Serverless | Pay per token · From $0.05 / 1M input tokens | Shared serverless capacity; usage draws down prepaid credits. Model and mode rates vary. |
| Reserved Throughput | Serverless rates, reserved by the minute | Guaranteed throughput on serverless and on-demand models; commitment required. |
| Fireworks Sovereign | Contract · Talk to us | Private preview for Enterprise; commitment required. |
| Serverless Training API | By model · see table below | Billed per 1M tokens prefilling, sampling, and training; rates vary by model. |
| Dedicated Training API | From $8.00 / GPU-hour | Billed per GPU-hour at the listed GPU rates. |
| We build with you | Priced per engagement. | The vendor describes forward-deployed researchers and engineers working with the customer team. |
| Pro | $8 / month | For individual developers; usage is against the plan cap. |
| Team | $10 per seat / month | For teams, with usage per member; up to 200 seats. |
| Enterprise | Contact sales | Custom terms and controls; listed features include SSO, SCIM, and audit logs. |
Checked 18 Nov 2026 · source
- Free plan
- The pricing page says Muse Glimmer 30B SFT is free through November 30 in the UI and Serverless Training API, with no credit card required. · pricing page
- Free trial
- Not stated (checked 18 Nov 2026)
- Trains on customer data
- The vendor says it does not use prompts, training data, or API inputs to train or improve its models without explicit opt-in. · privacy page
- Data retention
- The vendor says zero data retention is the default; it says prompt and generation data for open models are not logged or stored without explicit user opt-in. · pricing page, privacy page
- Data regions
- The vendor says one endpoint spans 30+ regions and supports policy-aligned data residency. · home page
- Data processing agreement
- Not stated (checked 18 Nov 2026)
- AI models used
- The vendor says its catalog has 100+ text, vision, audio, image, and embedding models. Named examples include Kimi K3, DeepSeek V4.1 Flash, and MiniMax M3. · pricing page, docs
- API
- The docs describe serverless inference API calls; the pricing page also lists Serverless Training API and Dedicated Training API offerings. · pricing page, docs
- SDKs
- The changelog references the Fireworks Python SDK, released separately as fireworks-ai, and a Python training cookbook. · changelog
- Webhooks
- Not stated (checked 18 Nov 2026)
- Platforms
- The vendor lists shared serverless capacity, dedicated GPU deployments, and Fireworks-managed inference in a customer's cloud account or data center. · pricing page, docs
- Security claims
- Vendor states SOC 2 Type II. · pricing pageVendor states ISO 27001, ISO 27701, and ISO 42001. · pricing pageThe vendor says HIPAA with a BAA is available. · pricing page
What we measured
- Package downloads · 30 days
- 1,634,751
- Tranco traffic rank · list of the day
- #31,480
How we tested · sources
- Hourly uptime and response-time checks of the product's URL.
- Repository, release and download data from GitHub, npm, PyPI and Hugging Face where they exist.
- We did not sign up, pay, or use the product as a customer.
- home page: https://fireworks.ai/ · fetched 18 Nov 2026
- pricing page: https://fireworks.ai/pricing · fetched 18 Nov 2026
- docs: https://docs.fireworks.ai/ · fetched 18 Nov 2026
- changelog: https://docs.fireworks.ai/updates/changelog · fetched 18 Nov 2026
- privacy page: https://fireworks.ai/privacy-policy · fetched 18 Nov 2026
- terms: https://cdn.sanity.io/files/pv37i0yn/production/60909f1a2f0cae74deb6ba7fc0f6eda8ab3bac4b.pdf · fetched 18 Nov 2026
- security page: https://trust.fireworks.ai/ · fetched 18 Nov 2026
Something wrong? Report an error with a public link that shows it; corrections re-run the review.
Top alternatives to Fireworks AI
See all- 79.3ToolFreemium · from $9/mo
- 77.5
API for building with Claude models.
ToolPaid - 76.1
Proxy and SDK to call 100+ LLM APIs in the OpenAI format.
ToolOpen source - 75.7
Google Cloud's platform to build with Gemini and other models.
ToolPaid