Comparison

Fireworks AI vs Together AI vs Amazon Bedrock

A side-by-side look at three managed platforms for running models, drawn from each vendor's own pricing and documentation pages.

At a glance

Three ways to serve models. Fireworks AI and Together AI sell per-token serverless inference with GPU-seconds or GPU-hours for dedicated capacity, while Amazon Bedrock sells per-token model access inside your own AWS account.

Quick answer

Identical on the record: Business model, Licence, Where it runs, Works offline, Runs on, Launched.

Fireworks AI vs Together AI vs Amazon Bedrock
Dimension	Fireworks AI	Together AI	Amazon Bedrock
Company	Fireworks AI	Together AI	Amazon Web Services
Category	Platforms & Infrastructure	Platforms & Infrastructure	Platforms & Infrastructure
Business model	Paid	Paid	Paid
Licence	—	—	—
Pricing	Pricing is usage-based per token, with dedicated capacity options; new developers receive free credits. Pricing ↗	Pricing is usage-based: per-token for serverless endpoints and per GPU hour for dedicated capacity; free credits are offered for new users. Pricing ↗	Pricing is usage-based per input/output token, with on-demand and provisioned-throughput options. Pricing ↗
Documentation	Documentation ↗	Documentation ↗	Documentation ↗
Launched	2023	2023	2023-09
Availability	Global	Global	Global
Billing unit	Per token (serverless); per GPU second (on-demand deployments) Source ↗ Verified 2026-10-12	Per token (serverless); per GPU hour (dedicated inference and GPU clusters) Source ↗ Verified 2026-10-12	Per token (on-demand inference); Provisioned Throughput sold in model units Source ↗ Verified 2026-10-12
Committed capacity	On-demand GPUs from $8.00 per hour (H100 80 GB), billed per second Source ↗ Verified 2026-10-12	Dedicated GPUs from $3.99 per hour (HGX H100), with reserved terms Source ↗ Verified 2026-10-12	Provisioned Throughput reserved by model unit; pricing agreed with your AWS account team Source ↗ Verified 2026-10-12
Fine-tuning	Supervised, preference (DPO) and reinforcement fine-tuning, priced per 1M training tokens Source ↗ Verified 2026-10-12	LoRA and full fine-tuning priced per 1M tokens, with a per-model minimum charge Source ↗ Verified 2026-10-12	Model customisation priced per 1M training tokens plus a monthly custom-model storage fee Source ↗ Verified 2026-10-12
Batch inference	Asynchronous batch inference at scale Source ↗ Verified 2026-10-12	Batch API priced per 1M tokens Source ↗ Verified 2026-10-12	Batch inference at 50% below on-demand pricing Source ↗ Verified 2026-10-12
API compatibility	OpenAI-compatible; presented as a drop-in replacement for OpenAI Source ↗ Verified 2026-10-12	OpenAI-compatible API Source ↗ Verified 2026-10-12	Native Amazon Bedrock API through the AWS SDKs Source ↗ Verified 2026-10-12
Updated	2026-09-10	2026-10-09	2026-10-09

Fields taken from the product records use the record dates shown. Sourced claims are labelled per product: verified against the linked page on the date shown, taken from the product record without re-checking, or not independently verified.

When to choose each

Fireworks AI — When you want per-token serverless inference with an OpenAI-compatible API and per-second GPU deployments.
Together AI — When you want serverless inference, fine-tuning and rentable GPU clusters from one platform.
Amazon Bedrock — When you want model access inside your existing AWS account, with provisioned throughput and AWS-native controls.

Each tool on the record

Fireworks AI

Fireworks AI is made by Fireworks AI. The record lists it as Paid, first recorded in 2023, and fit for Running models yourself. Its values below carry their own source and date.

Fireworks AI is listed under Platforms & Infrastructure: Hubs and infrastructure for hosting, sharing, and running AI models and datasets.

Company: Fireworks AI
Category: Platforms & Infrastructure
Business model: Paid
Licence: proprietary
Launched: 2023
Availability: Global
Tasks it fits: Running models yourself
How you use it: Web app
Runs on: Web
Where it runs: Vendor cloud
Works offline: Not recorded
Commercial use: Not recorded

Pricing on the record: Pricing is usage-based per token, with dedicated capacity options; new developers receive free credits.

Sources on the record: Official website, Documentation, Official pricing.

Not recorded for Fireworks AI: Works offline, Commercial use. An empty field is reported as unknown rather than guessed.

Together AI

Together AI is made by Together AI. The record lists it as Paid, first recorded in 2023, and fit for Running models yourself. Its values below carry their own source and date.

Together AI is listed under Platforms & Infrastructure: Hubs and infrastructure for hosting, sharing, and running AI models and datasets.

Company: Together AI
Category: Platforms & Infrastructure
Business model: Paid
Licence: proprietary
Launched: 2023
Availability: Global
Tasks it fits: Running models yourself
How you use it: Web app
Runs on: Web
Where it runs: Vendor cloud
Works offline: Not recorded
Commercial use: Not recorded

Pricing on the record: Pricing is usage-based: per-token for serverless endpoints and per GPU hour for dedicated capacity; free credits are offered for new users.

Sources on the record: Official website, Documentation, Overview - Together AI docs, Official pricing.

Not recorded for Together AI: Works offline, Commercial use. An empty field is reported as unknown rather than guessed.

Amazon Bedrock

Amazon Bedrock is made by Amazon Web Services. The record lists it as Paid, first recorded in 2023, and fit for Running models yourself. Its values below carry their own source and date.

Amazon Bedrock is listed under Platforms & Infrastructure: Hubs and infrastructure for hosting, sharing, and running AI models and datasets.

Company: Amazon Web Services
Category: Platforms & Infrastructure
Business model: Paid
Licence: proprietary
Launched: 2023-09
Availability: Global
Tasks it fits: Running models yourself
How you use it: Web app
Runs on: Web
Where it runs: Vendor cloud
Works offline: Not recorded
Commercial use: Not recorded

Pricing on the record: Pricing is usage-based per input/output token, with on-demand and provisioned-throughput options.

Sources on the record: Official website, Documentation, Official pricing.

Not recorded for Amazon Bedrock: Works offline, Commercial use. An empty field is reported as unknown rather than guessed.

Frequently asked questions

Which of these has been available longest?

Fireworks AI — first recorded in 2023.

Which of these is open source?

None of these is open source; every tool in this comparison is proprietary.

Which of these works offline?

None of these is recorded as working offline.

Which of these has a public API?

A public API is recorded for Fireworks AI, Together AI, Amazon Bedrock.

Which platforms do these run on?

Recorded platforms — Fireworks AI: Web · Together AI: Web · Amazon Bedrock: Web.

Can any of these be self-hosted or run locally?

None of these is recorded as self-hosted or local; all run as a hosted service.

Can the per-token prices be compared directly?

No. Each platform prices different models, and the same model can sit in several tiers, so the unit is comparable but the numbers are not.

Do all three offer committed capacity?

Yes: Fireworks AI and Together AI rent GPUs by the second or hour, and Amazon Bedrock sells reserved Provisioned Throughput in model units.

Is the API OpenAI-compatible?

Fireworks AI and Together AI are; Amazon Bedrock uses its own API, reached through the AWS SDKs.

How to use a comparison

Start from the task, not the brand: filter the index by what you need to get done.
Check the evidence: a value marked “verified” names the source page and the date it was checked.
Prefer the vendor’s own page for the final decision: prices and limits change without notice.

Related guides

AI platforms and infrastructure

Sources and record dates

Record dates describe index updates, not independent product tests. Check linked documentation for current features and pricing.