# Baseten vs Fireworks AI

Canonical: https://inetgeek.com/compare/baseten-vs-fireworks/

Every value below is read from Baseten's and Fireworks AI's own documentation. See https://inetgeek.com/methodology/ for how.

## At a glance

| Criterion | Baseten | Fireworks AI |
| --- | --- | --- |
| Pricing model | No platform fee: per-token Model APIs for open models, and dedicated deployments billed per minute of compute with volume discounts. The Pro tier buys priority access to high-demand GPUs rather than a lower rate. | Three products on one bill. Serverless inference is per token, per model. Fine-tuning is per million TRAINING tokens by parameter band — $0.50 for models up to 16B, $10.00 above 300B. Dedicated GPUs are hourly: $7.00 for an H100 or H200, $10.00 for a B200. |
| Input price, top model | $1.40 per million input tokens for GLM-5.3, with cached input at $0.14 and output at $4.40. GLM-5.2 Fast, a speed variant, is $2.10. | $3.00 per million input tokens for Kimi K3, the most expensive model in its serverless catalogue. Fireworks hosts other labs' open-weight models rather than its own, so "flagship" here means the top of what it serves. |
| Output price, top model | $4.40 per million output tokens for GLM-5.3. | $15.00 per million output tokens for Kimi K3. |
| Input price, cheapest model | $0.10 per million input tokens for GPT OSS 120B, with output at $0.50. GLM-5.3-Flash is $0.15 in and $0.50 out. | $0.15 per million input tokens for GLM 5.3 Flash, with output at $0.50. DeepSeek-V4-Flash is $0.22 in and $0.66 out — the same model DeepSeek itself sells at $0.44 peak. |
| Context window | 1,048K tokens on the DeepSeek V4 line and the GLM 5.2/5.3 family, the widest it serves; 262K on the Kimi K2 models, 200K on GLM 4.7 and 128K on GPT OSS 120B. Baseten's table publishes these in thousands, so the magnitude is 1,048,000 rather than the 1,048,576 a power-of-two reading would give — the vendor's own figure, not a conversion of it. | 1,048,576 tokens on Kimi K3, GLM 5.3 and both DeepSeek V4 variants; 262,144 on Qwen3.8 and 131,072 on Muse Glimmer 30B. The window is the model's, not Fireworks'. |
| Max output tokens | 262K tokens on most models, and it is a separate ceiling from the context window rather than the remainder of it: DeepSeek V4 Flash 0731 allows 384K output inside a 1,048K window, while GLM 4.7, Nemotron Ultra and GPT OSS 120B cap output at their full context. The Inkling models are the outlier at 32K. | No separate output cap. max_tokens is bounded by the model's context length — prompt plus completion must fit inside it — and what happens on overflow is set by context_length_exceeded_behavior rather than a fixed ceiling. |
| Prompt caching | Supported | Supported |
| H100 SXM, per GPU-hour | $6.50 per hour for a dedicated H100 80GB. Baseten publishes $0.10833 per MINUTE — the hourly figure is that times 60, ours rather than Baseten's, and the page offers an hourly toggle. A100 80GB is $0.06667 a minute and B200 $0.16633. | $7.00 per hour for an H100 80GB on demand, or $8.00 on the shorter commitment shown beside it. The same rate applies to an H200 141GB. |
| Largest GPU offered | 180GB per GPU on a B200, at $0.16633 per minute — $9.98 an hour. | 180GB per GPU on a B200, at $10.00 an hour on demand. |

## Where they differ

### Pricing model

- Baseten: No platform fee: per-token Model APIs for open models, and dedicated deployments billed per minute of compute with volume discounts. The Pro tier buys priority access to high-demand GPUs rather than a lower rate. ([source](https://www.baseten.co/pricing/))
- Fireworks AI: Three products on one bill. Serverless inference is per token, per model. Fine-tuning is per million TRAINING tokens by parameter band — $0.50 for models up to 16B, $10.00 above 300B. Dedicated GPUs are hourly: $7.00 for an H100 or H200, $10.00 for a B200. ([source](https://fireworks.ai/pricing))

### Input price, top model

- Baseten: $1.40 per million input tokens for GLM-5.3, with cached input at $0.14 and output at $4.40. GLM-5.2 Fast, a speed variant, is $2.10. ([source](https://www.baseten.co/pricing/))
- Fireworks AI: $3.00 per million input tokens for Kimi K3, the most expensive model in its serverless catalogue. Fireworks hosts other labs' open-weight models rather than its own, so "flagship" here means the top of what it serves. ([source](https://fireworks.ai/models))

### Output price, top model

- Baseten: $4.40 per million output tokens for GLM-5.3. ([source](https://www.baseten.co/pricing/))
- Fireworks AI: $15.00 per million output tokens for Kimi K3. ([source](https://fireworks.ai/models))

### Input price, cheapest model

- Baseten: $0.10 per million input tokens for GPT OSS 120B, with output at $0.50. GLM-5.3-Flash is $0.15 in and $0.50 out. ([source](https://www.baseten.co/pricing/))
- Fireworks AI: $0.15 per million input tokens for GLM 5.3 Flash, with output at $0.50. DeepSeek-V4-Flash is $0.22 in and $0.66 out — the same model DeepSeek itself sells at $0.44 peak. ([source](https://fireworks.ai/models))

### Context window

- Baseten: 1,048K tokens on the DeepSeek V4 line and the GLM 5.2/5.3 family, the widest it serves; 262K on the Kimi K2 models, 200K on GLM 4.7 and 128K on GPT OSS 120B. Baseten's table publishes these in thousands, so the magnitude is 1,048,000 rather than the 1,048,576 a power-of-two reading would give — the vendor's own figure, not a conversion of it. ([source](https://docs.baseten.co/inference/model-apis/overview))
- Fireworks AI: 1,048,576 tokens on Kimi K3, GLM 5.3 and both DeepSeek V4 variants; 262,144 on Qwen3.8 and 131,072 on Muse Glimmer 30B. The window is the model's, not Fireworks'. ([source](https://fireworks.ai/models))

### Max output tokens

- Baseten: 262K tokens on most models, and it is a separate ceiling from the context window rather than the remainder of it: DeepSeek V4 Flash 0731 allows 384K output inside a 1,048K window, while GLM 4.7, Nemotron Ultra and GPT OSS 120B cap output at their full context. The Inkling models are the outlier at 32K. ([source](https://docs.baseten.co/inference/model-apis/overview))
- Fireworks AI: No separate output cap. max_tokens is bounded by the model's context length — prompt plus completion must fit inside it — and what happens on overflow is set by context_length_exceeded_behavior rather than a fixed ceiling. ([source](https://fireworks.ai/docs/api-reference/post-chatcompletions))

### H100 SXM, per GPU-hour

- Baseten: $6.50 per hour for a dedicated H100 80GB. Baseten publishes $0.10833 per MINUTE — the hourly figure is that times 60, ours rather than Baseten's, and the page offers an hourly toggle. A100 80GB is $0.06667 a minute and B200 $0.16633. ([source](https://www.baseten.co/pricing/))
- Fireworks AI: $7.00 per hour for an H100 80GB on demand, or $8.00 on the shorter commitment shown beside it. The same rate applies to an H200 141GB. ([source](https://fireworks.ai/pricing))

### Largest GPU offered

- Baseten: 180GB per GPU on a B200, at $0.16633 per minute — $9.98 an hour. ([source](https://www.baseten.co/pricing/))
- Fireworks AI: 180GB per GPU on a B200, at $10.00 an hour on demand. ([source](https://fireworks.ai/pricing))

## Which should you choose?

Pick Baseten if Teams that want per-token APIs and dedicated GPUs from one vendor with no plan fee — minute-granularity billing means a model that runs for ten minutes costs ten minutes.

Pick Fireworks AI if Running open-weight models without operating GPUs — the same DeepSeek and Qwen checkpoints the labs publish, served per token, with a path to dedicated hardware if throughput demands it.

Consider something else: Baseten — Its dedicated H100 works out at $6.50 an hour, roughly three times DeepInfra's, so steady GPU workloads pay a premium for the platform around them.

Consider something else: Fireworks AI — You want a frontier proprietary model: Fireworks serves other people's open weights, so the ceiling is whatever the open ecosystem has released.

## Documented by only one

- Free tier: Fireworks AI: Yes
- Entry paid plan: Baseten: $0 per month on the Basic plan — pay as you go, with dedicated deployments, model APIs and training included rather than gated behind a fee.
- Rate limits: Baseten: Two limits, requests and tokens per minute, set by account status rather than spend: an unverified Basic account gets 15 RPM and 100,000 TPM, a verified Basic or Pro account 120 RPM and 500,000 TPM, Enterprise custom. Cached input counts toward TPM at full weight even though it is billed cheaper.

## Questions this comparison answers

**Should I choose Baseten or Fireworks AI?**

Pick Baseten if Teams that want per-token APIs and dedicated GPUs from one vendor with no plan fee — minute-granularity billing means a model that runs for ten minutes costs ten minutes.
Pick Fireworks AI if Running open-weight models without operating GPUs — the same DeepSeek and Qwen checkpoints the labs publish, served per token, with a path to dedicated hardware if throughput demands it.