Fireworks AI

Last updated: Nov 4, 2026

Rationale

Fireworks AI is the inference provider of Fluid Attacks for open source (open-weight) models. With it, Fluid Attacks can experiment with models from more than one lab and use them in production.

These are the primary reasons why Fluid Attacks selected it and not other alternatives:

It adds the most recent models from important labs much faster than Amazon Bedrock. Bedrock usually takes months to make a newly released open source model available, although Bedrock gives open source models in theory.
In the tests of Fluid Attacks, it was stable and consistent under continuous load, but the alternatives that Fluid Attacks evaluated were not (read the Alternatives section below).
It supports the request concurrency (requests per minute) that Fluid Attacks must have for experimentation and production workloads.
It gives Zero Data Retention (ZDR) by default.
It has these compliance certifications: SOC 2 Type II, HIPAA, GDPR, ISO 27001:2022, ISO 27701, and ISO/IEC 42001:2023.
Its infrastructure is in the US.

Alternatives

Fluid Attacks examined these alternatives, but it did not select them for these reasons:

DeepInfra

It was not consistent under load. At a low target of 10 requests per minute, a pilot run completed successfully only 70% of the requests. The mean latency was more than 30 seconds, and timeouts occurred frequently. Thus, Fluid Attacks stopped the test.
At a higher target of 1,000 requests per minute, a complete load run had better results, and it was even faster than Fireworks AI. But the inconsistency at lower concurrency caused doubts about its general reliability.

Baseten

Its basic account tier supports only a small number of requests per minute, much less than the concurrency that Fluid Attacks must have. Thus, Fluid Attacks could not even do a load test at its target scale.

Amazon Bedrock

Open source models are available in theory, but Bedrock usually takes months to add the most recent models from important labs. This is not as fast as Fluid Attacks must be to use new open source releases.

Fluid Attacks uses Fireworks AI as an inference provider for open source models. It runs together with Amazon Bedrock, Anthropic, and OpenAI, which continue to be the providers of Fluid Attacks for proprietary models. Each AI feature of Fluid Attacks can use Fireworks AI for open source model inference if an internal evaluation of that feature approves it. This applies to experimentation and production loads.