Your cookie choices

We use necessary cookies to make this site work. With your permission, analytics cookies help us understand usage and marketing cookies support advertising.

You can reject non-essential cookies and still use the site. Change your choices anytime in Cookie settings.

Cookie settingsCookie policy

Fireworks AI review

Updated September 28, 2026 · DeployCue Editorial Team · Facts checked September 26, 2026

We earn a commission when you sign up through links on this site. Partner status can affect where a brand appears in our rankings. Partners with a live offer in your country come first, then every other brand by score.

Our verdict

Fireworks AI suits developers who want many model families behind a single OpenAI-compatible endpoint, and a bill that counts tokens rather than plan tiers. Its overall score of 7.3 pairs a strong offer with a cost mark nearer the middle of the field.

Key facts

Best for
Teams wanting a single OpenAI-compatible endpoint for many open models on a per-token bill.
Input price from
$0.050↗
Output price from
$0.20↗
Models offered
Kimi K3, Kimi K2.7 Code, Kimi K2.6, DeepSeek V4.1 Flash, DeepSeek V4 Pro (0813), DeepSeek V4 Flash (0731), GLM 5.3, GLM 5.3 Flash, GLM 5.2, Qwen 3.8 Max, MiniMax M3, OpenAI GPT OSS 120B, Muse Glimmer 30B, NVIDIA Nemotron 3.5 Lightning 30B A3B, NVIDIA Nemotron 3 Ultra, Qwen3 Embedding 8B, Qwen3 Reranker 8B, Voyage 4, Voyage 4 Lite, Voyage ReRank 2.5, BGE-M3, FLUX.1 [schnell]↗
Largest context window
1,048,576↗
Free credits
$1 in free credits↗
OpenAI-compatible API
Yes↗
Compliance certifications
HIPAA, SOC 2 Type II, GDPR↗
Legal entity
Fireworks AI, Inc.↗
Headquarters
United States↗
Founded
2022↗
Support channels
Help-centre, Community↗
Official website
fireworks.ai

LLM gateways

Fee model
Usage-based: per token for serverless inference and fine-tuning, per GPU second/hour for on-demand and dedicated deployments↗
Bring your own key
Yes↗

Facts checked September 26, 2026, 8 sources

Apps and profiles

Elsewhere
LinkedInXdiscord.comyoutube.com

Pricing

  • Fireworks AI Qwen3 235B A22B (USD)

    input-tokens: $0.22 / million

  • Fireworks AI Qwen3 235B A22B (USD)

    output-tokens: $0.88 / million

  • Fireworks AI MiniMax M2 (USD)

    input-tokens: $0.30 / million

  • Fireworks AI MiniMax M2 (USD)

    output-tokens: $1.20 / million

  • Fireworks AI MiniMax M2 (USD)

    cached-input-tokens: $0.06 / million

  • Fireworks AI Llama 4 Maverick (USD)

    input-tokens: $0.22 / million

  • Fireworks AI Llama 4 Maverick (USD)

    output-tokens: $0.88 / million

  • Fireworks AI Llama 3.1 8B (USD)

    input-tokens: $0.20 / million

  • Fireworks AI Llama 3.1 8B (USD)

    output-tokens: $0.20 / million

  • Fireworks AI Kimi K2 (USD)

    input-tokens: $0.60 / million

  • Fireworks AI Kimi K2 (USD)

    output-tokens: $3.00 / million

  • Fireworks AI Kimi K2 (USD)

    cached-input-tokens: $0.10 / million

  • Fireworks AI gpt-oss-20b (USD)

    input-tokens: $0.07 / million

  • Fireworks AI gpt-oss-20b (USD)

    output-tokens: $0.30 / million

  • Fireworks AI gpt-oss-20b (USD)

    cached-input-tokens: $0.04 / million

  • Fireworks AI gpt-oss-120b (USD)

    input-tokens: $0.15 / million

  • Fireworks AI gpt-oss-120b (USD)

    output-tokens: $0.60 / million

  • Fireworks AI gpt-oss-120b (USD)

    cached-input-tokens: $0.02 / million

  • Fireworks AI Gemma 3 27B (USD)

    input-tokens: $0.10 / million

  • Fireworks AI Gemma 3 27B (USD)

    output-tokens: $0.10 / million

  • Fireworks AI DeepSeek V3 (USD)

    input-tokens: $0.56 / million

  • Fireworks AI DeepSeek V3 (USD)

    output-tokens: $0.90 / million

  • Fireworks AI DeepSeek V3 (USD)

    cached-input-tokens: $0.28 / million

  • Fireworks AI DeepSeek R1 (USD)

    input-tokens: $3.00 / million

  • Fireworks AI DeepSeek R1 (USD)

    output-tokens: $8.00 / million

Pros and cons

Pros

  • Serverless inference and fine-tuning are priced per token, from $0.05 for input and $0.20 for output.
  • The widest context window reaches 1,048,576 tokens and ranks 11 of 72 on this site.
  • Bring your own key is supported and the API is OpenAI-compatible, so a team can keep its own credentials and its existing client.
  • Free credits of $1 in free credits are on offer, so a trial integration can run before real spend starts.
  • The published compliance set is HIPAA, SOC 2 Type II and GDPR, which is what regulated buyers check first.

Cons

  • Support runs on help centre and community only, with no live agent channel listed.
  • Fireworks AI ranks 28 of 71 on lowest input price and 32 of 70 on lowest output price, so entry pricing sits in the middle of the field.
  • The usage-based fee model pairs token rates from $0.05 with GPU-second and GPU-hour billing for on-demand and dedicated deployments, so forecasts need both.
  • Azure AI Foundry lists gpt 6 astra, gpt 6 luna and gpt 6 sol, a first-party line the Kimi K3, Kimi K2.7 Code and Kimi K2.6 catalog does not carry.
Fireworks AI

Ready to try Fireworks AI?

Scored 9.1 / 10 by DeployCue.

See alternatives

Who it suits

  • Developers who want a single OpenAI-compatible endpoint across the Kimi, DeepSeek, GLM and Qwen lines.
  • Small teams moving a prototype into production with $1 in free credits to start.
  • Enterprise buyers who need a published compliance set: HIPAA, SOC 2 Type II and GDPR.

Who should look elsewhere

  • Salad, if the job is a routing gateway rather than a hosted catalog, since Salad starts at $0.06.
  • Azure AI Foundry, if you want a first-party cloud catalog such as gpt 6 astra, gpt 6 luna and gpt 6 sol under one contract.
  • Supascale AI, if breadth of families across vendors matters more than a per-token host, given the range in DeepSeek V4 Pro, GPT 5.6 Sol and GPT 5.6 Terra.

Where it sits among rivals

On the rankings here, Fireworks AI places 28 of 71 brands on lowest input price and 32 of 70 on lowest output price. Its largest context window ranks 11 of 72. The pattern is consistent: capacity sits higher up the field than price does, so a reader choosing on context length gets a better position here than a reader choosing purely on the cheapest token rate. Nobody plans around a ranking alone, but it sets the starting expectation for where the entry price lands against rivals selling inference at similar levels.

Models and context

The catalog runs from Kimi K2.7 Code and DeepSeek through GLM 5.3 and Qwen 3.8 Max, with embedding and reranking entries such as Qwen3 Embedding 8B, Voyage 4 and BGE-M3, plus an image model in FLUX.1 [schnell]. The widest context window on offer is 1,048,576 tokens, which is enough to hold large documents or a codebase in a single pass rather than a sliding window. Calls go out over an OpenAI-compatible API, so the model name is what changes between experiments, not the shape of the request.

What it connects to

An OpenAI-compatible API is the main integration surface, and bring your own key is supported for teams that hold provider credentials themselves. That pairing keeps existing clients pointed at the same request and response shapes while credentials stay with the team that already manages them. It also suits a mixed estate, where different projects land on different families in the same catalog and the same client code calls each one.

Who runs it

The vendor behind the platform is Fireworks AI, Inc., headquartered in the US and founded in 2022. Support runs through help centre and community, so self-service documentation and community answers carry most questions, and account or contract work is worth raising directly. The contracting entity is Fireworks AI, Inc., the same name that appears on pricing, which is the one to put on paperwork and data processing terms.

How Fireworks AI compares

Frequently asked questions

What will a first bill look like?+

Serverless inference and fine-tuning are priced per token, starting at $0.05 for input and $0.20 for output. On-demand and dedicated deployments bill by the GPU second or hour instead, so the deployment style decides which rate applies to a given run.

Is there a way to try it without spending?+

New accounts carry $1 in free credits, and serverless calls run at $0.05 for input tokens once that is spent, so a small test costs little before a contract. Any model in Kimi K3, Kimi K2.7 Code and Kimi K2.6 can serve that first run.

How much context fits in one request?+

The widest window listed is 1,048,576 tokens, so long documents and codebases can go in a single pass. Individual models differ, so check the window on the exact model you plan to call before designing around it.

Which models are on the menu?+

The catalog covers text, embedding, reranking and image work, from Kimi K2.7 Code and DeepSeek through GLM 5.3, Qwen 3.8 Max and Voyage 4 to BGE-M3 and FLUX.1 [schnell]. Treat Kimi K3, Kimi K2.7 Code and Kimi K2.6 as the live list rather than a summary.

Can I use my own provider keys?+

Bring your own key is supported, and the API is OpenAI-compatible, so a team can hold its own provider credentials and still call models such as DeepSeek or GLM 5.3 through the same client. Without your own key, usage bills to the account you open, with serverless calls from $0.05 for input.

Does it meet compliance requirements?+

The published compliance set is HIPAA, SOC 2 Type II and GDPR, which covers the frameworks enterprise buyers usually ask about first. Contracts and data handling terms sit with Fireworks AI, Inc., so check them against your own obligations before signing.

Who handles support questions?+

Support channels are help centre and community, so most questions get answered by documentation and peers rather than a live agent. Account and contract work is better handled by going straight to the company behind the platform, Fireworks AI, Inc..

Who is Fireworks AI for?+

It suits developers and small teams that want a wide open-model catalog on a per-token bill, with $1 in free credits to try it and $0.05 rates to keep paying. Teams that need a single cloud catalog under one contract may prefer Azure AI Foundry.

Sources

  • docs.fireworks.ai/guides/querying-text-models checked September 26, 2026

    Backs: OpenAI-compatible API

    “Fireworks provides fast, cost-effective access to leading open-source text models through OpenAI-compatible APIs.”

  • docs.fireworks.ai/serverless/pricing checked September 26, 2026

    Backs: Input price from, Output price from

    “NVIDIA Nemotron 3.5 Lightning 30B A3B | $0.05 / $0.01 / $0.20”

  • en.wikipedia.org/wiki/Fireworks_AI checked September 26, 2026

    Backs: Founded, Headquarters

    “Established in 2022, the company was founded by a team of engineers from Meta's artificial intelligence and machine learning groups.”

  • fireworks.ai/contact checked September 26, 2026

    Backs: Support channels

    “Product support, bugs, and account issues are handled in Discord and Docs”

  • fireworks.ai/enterprise checked September 26, 2026

    Backs: Compliance certifications

    “Fireworks is fully HIPAA, SOC2-type2, and GDPR compliant.”

  • fireworks.ai/models checked September 26, 2026

    Backs: Largest context window, Models offered

    “Ember-1 $3/M Input - $15/M Output - 1048576 Context”

  • fireworks.ai/nexus checked September 26, 2026

    Backs: Bring your own key

    “Difficult tasks pass through to your closed provider on your key”

  • fireworks.ai/pricing checked September 26, 2026

    Backs: Fee model, Free credits, Legal entity

    “Pay per token, with high rate limits and postpaid billing”

Compiled from 8 source pages, checked September 26, 2026. We did not open an account.

Fireworks AI

Fireworks AI

Teams wanting a single OpenAI-compatible endpoint for many open models on a per-token bill.

9.1/10See alternatives