
Fireworks
AI video generation platform.
Pricing Change History
23 changes · 5 summariesKimi K3.1 Pro added to highlighted serverless rates: $2.00 input / $0.20 cached / $10.00 output per 1M tokens; priority $3.00 / $0.30 / $15.00
Includes Nov 14 · first seen Nov 14, 2026
New model Kimi K3.1 Pro added to Highlighted Serverless rates table at $2.00/1M input, $0.20/1M cached input, $10.00/1M output ($3.00/$0.30/$15.00 Priority)
Pricing page now displays SOC 2 Type 2, ISO 27001, ISO 27701, and ISO 42001 compliance certification badges, which were absent from the prior version
Serverless Training API adds two base models: Muse Spark Open ($6.19 prefill, $18.58 train per 1M) and DeepSeek V4.1 Flash ($0.60 prefill, $4.98 train per 1M).
Includes Oct 17 · first seen Oct 17, 2026
Serverless Training API table adds Muse Spark Open (256K; $6.19 prefill, $15.48 sample, $18.58 train) and DeepSeek V4.1 Flash (256K; $0.60 prefill, $3.60 sample, $4.98 train) per 1M tokens
Fireworks added GLM 5.3 Flash as a new base model in the Serverless Training API pricing table, with its own 200K context window and per-1M-token pricing for prefill, cached prefill, sample, and train.
Added GLM 5.3 model (262K context) to the Serverless Training API pricing table: $4.86 prefill/1M, $0.972 cached prefill/1M, $12.15 sample/1M, $14.58 train/1M tokens
Serverless Training API model catalog updated: Qwen 3.5 9B (64k, $0.66 prefill) removed and Qwen 3.6 27B replaced by Qwen 3.8 27B at identical pricing; DeepSeek V4 Flash 0731 (262K) and Muse Glimmer 30B (128K) added with per-token prefill/cached/sample/train pricing.
Includes Sep 1 · first seen Sep 1, 2026
Serverless Training API catalog updated: removed Qwen 3.5 9B, renamed Qwen 3.6 27B to Qwen 3.8 27B (same pricing), and added DeepSeek V4 Flash 0731 and Muse Glimmer 30B with new per-token pricing.
On-Demand GPU pricing rising from Sep 1: H100/H200 $7.00→$8.00, B200 $10.00→$13.00, B300 $12.00→$15.00, GB300 $18.00→$20.00/hr; region surcharge increased 1.25x→1.5x.
On-Demand Pricing added a new GB300 288GB GPU tier at $18.00/hour and a new note that region-restricted deployments (US, Europe) are priced at 1.25x the listed rates. Serverless Training API: Qwen 3.6 27B context window increased from 65k to 128k tokens (prefill/sample/train prices unchanged).
Added Serverless Training API: LoRA training on a shared always-on trainer pool for Qwen 3.5 9B, Qwen 3.6 27B, and Kimi K3, priced per 1M tokens for prefill/cached-prefill/sample/train.
Includes Jul 28 · first seen Jul 28, 2026
Fireworks launched a new Serverless Training API: pay-per-token LoRA training (no provisioning/idle cost) for Qwen 3.5 9B, Qwen 3.6 27B, and Kimi K3, priced per prefill/cached-prefill/sample/train tokens.
On-Demand GPU pricing increased: H100 $6.00→$7.00/hr, H200 $6.00→$7.00/hr, B200 $9.00→$10.00/hr, new B300 tier at $12.00/hr. Also, the detailed per-model serverless pricing table was replaced with a docs-page link.
Includes 4 weeklies · first seen Apr 6, 2026
Serverless pricing tier renamed from ‘Turbo’ to ‘Fast’. The ‘Turbo + Priority tiers’ description on the pricing page now reads ‘Fast + Priority tiers’.
On-Demand GPU prices increased as previously announced (May 1, 2026): H100 $6.00→$7.00/hr, H200 $6.00→$7.00/hr, B200 $9.00→$10.00/hr, B300 $11.00→$12.00/hr. Advance notice note removed. Serverless Text and Vision model prices removed from pricing page and redirected to docs.fireworks.ai/serverless/pricing. All per-model token prices (DeepSeek, Kimi, GLM, OpenAI gpt-oss, MiniMax) no longer publicly listed.
Speech to Text and Image Generation sections removed from pricing page; DeepSeek-V4-Pro added at $1.74/$0.145 cached/$3.48 per 1M tokens.
New On-Demand GPU tier B300 288 GB added to on-demand deployments pricing at $11.00/hour billed per second. Fine tuning pricing expanded from 2 columns (Supervised Fine Tuning, DPO) to 4 columns (LoRA SFT, LoRA DPO, Full Param SFT, Full Param DPO) with new Full Param prices added per model tier.
Added Kimi K2.5 Turbo model with pricing at $0.99 input, $0.16 cached input, $4.94 output per 1M tokens
Current plans
| Plan | Price |
|---|---|
| Serverless | Usage-based |
Also priced by usage
| |
| Fine Tuning | $0.50 per 1M training tokens, LoRA SFT (models up to 16B params) |
| Enterprise | Contact |
| Reserved Throughput | Contact |
| On-Demand | Usage-based |
Also priced by usage
| |
Company Facts
Pulse news about Fireworks
View all →How Fireworks compares with its peers
| Company | Entry price |
|---|---|
A AdobeLast change: Jun 22, 2026![]() | $4.99 /mobilled annually |
S SketchLast change: Nov 2, 2026![]() | $12 /editor/mobilled annually |
F FigmaLast change: Nov 13, 2026![]() | $5 /collab seat/mobilled annually |
A AffinityLast change: Nov 9, 2026![]() | Free |
C CorelLast change: Nov 9, 2026![]() | $79.99 /mobilled annually |
Fireworks pricing FAQ
- How much does Fireworks cost?
- Fireworks prices on usage: the Serverless plan is $0.08 per 10M input tokens (embeddings, up to 150M params) + $0.16 per 10M input tokens (embeddings, 150M-350M params).
- Does Fireworks have a free plan?
- No. Fireworks does not list a free tier.
- What pricing model does Fireworks use?
- Fireworks uses usage-based pricing across 5 plans.
- When did Fireworks last change its pricing?
- Nov 14, 2026: Fireworks lists Kimi K3.1 Pro on serverless at $2.00 input, $10.00 output per 1M.
- How often has Fireworks changed its pricing?
- Pulse has recorded 23 pricing changes for Fireworks since it started tracking on Nov 17, 2024; the pricing page was last checked on Nov 20, 2026.
- What does “tracked since” mean?
- It's the date of the first saved copy of the pricing page that Pulse checked for changes against a later copy. For a few companies we hold older copies that Pulse never checked for changes, and those don't change this date. If Pulse has never checked two copies, it's the date of the first saved copy.




