Fireworks hikes H100 GPU pricing 50% while trimming AI model lineup
What happened
Between the Dec 29, 2025 and Mar 30, 2026 captures of Fireworks' pricing page, Pulse recorded 3 changes: Streaming ASR v1/v2 removed from STT pricing. DeepSeek V3, Kimi K2 Instruct/Thinking, and gpt-oss-20b lost cached pricing. Caching footnote updated to 50% universal rate; DeepSeek R1 0528 ($1.35 input/$0.68 cached/$5.4 output), Qwen3 235B Family ($0.22 input/$0.88 output), and Qwen3 Coder 480B ($0.45 input/$0.23 cached/$1.80 output) removed from serverless pricing table; H100 80 GB GPU on-demand price increased from $4.00/hr to $6.00/hr (50% increase). MiniMax M2 family cached token price decreased from $0.15 to $0.03 per 1M tokens. Pulse files it as a pricing change (feature removed, price increase) with an upward direction. Pulse's latest capture (Nov 18, 2026) lists Serverless at $0.08 per 10M input tokens (embeddings, up to 150M params) + $0.16 per 10M input tokens (embeddings, 150M-350M params), Fine Tuning at $0.50 per 1M training tokens, LoRA SFT (models up to 16B params) and On-Demand at $13 per GPU hour (B200 180GB) + $15 per GPU hour (B300 288GB).
This quarterly roll-up includes 4 changes Pulse first recorded in a shorter period: "Fireworks Adds Kimi K2.5 Turbo at $0.99 per Million Input Tokens" (week of Mar 9, 2026, saved Mar 16, 2026), "Fireworks clarifies fine-tuning with reasoning traces increases token usage" (week of Feb 23, 2026, saved Mar 2, 2026) and "Fireworks overhauls AI model lineup with GLM-5 and MiniMax M2 additions" (week of Feb 9, 2026, saved Feb 18, 2026) and 1 more. It also records changes none of those periods did (price increase), so Pulse counts it as a change of its own. Before that, Fireworks' previous recorded change was "Fireworks Pricing Restructured: DeepSeek R1 0528 Sees 55% Input and 32.5% Output Price Reduction" (Jan 10, 2026). Fireworks has been tracked since 2024-11-17.
In the same window Sketch, Adobe and Figma also changed their Marketing & Content pricing pages. Those moves are listed further down, and the Pulse Agent can compare them side by side.
- RemovedSTTFeatureStreaming ASR v1/v2Not listed
- Feature removedServerlessFeature
DeepSeek R1 0528 ($1.35 input/$0.68 cached/$5.4 output), Qwen3 235B Family ($0.22 input/$0.88 output), and Qwen3 Coder 480B ($0.45 input/$0.23 cached/$1.80 output) removed from serverless pricing table.
- +50%On-DemandPrice$4.00/hr$6.00/hr
| Plan | Field | Before | After | Change |
|---|---|---|---|---|
| STT | Feature | Streaming ASR v1/v2 | Not listed | Removed |
| Serverless | Feature | DeepSeek R1 0528 ($1.35 input/$0.68 cached/$5.4 output), Qwen3 235B Family ($0.22 input/$0.88 output), and Qwen3 Coder 480B ($0.45 input/$0.23 cached/$1.80 output) removed from serverless pricing table. | Feature removed | |
| On-Demand | Price | $4.00/hr | $6.00/hr | +50% |
Values as written on Fireworks' public pricing page in the two captures Pulse compared; where the page states no number, the cells give the wording before and after.
Evidence
The two captures Pulse comparedFull-page captures of Fireworks' pricing page (fireworks.ai) from Dec 29, 2025 and Mar 30, 2026; the change is marked around the middle of the page, where on-demand price reads $4.00/hr and then $6.00/hr.
Sign in to see the evidence
The two full-page captures with this change marked up are free with an account.
Fireworks, the last 90 days
Full history- May 12, 2026Price increaseFireworks hikes GPU prices, moves model pricing behind docs wall
- May 4, 2026PackagingFireworks adds DeepSeek-V4-Pro at $3.48/1M tokens, drops speech and image tiers
- Apr 9, 2026PackagingFireworks expands fine-tuning tiers and adds B300 GPU deployments
- Apr 3, 2026Price increaseFireworks hikes H100 GPU pricing 50% while trimming AI model lineup This change
- Mar 16, 2026ImprovementFireworks Adds Kimi K2.5 Turbo at $0.99 per Million Input Tokens
- Mar 3, 2026ImprovementFireworks clarifies fine-tuning with reasoning traces increases token usage
- Feb 18, 2026ImprovementFireworks overhauls AI model lineup with GLM-5 and MiniMax M2 additions
- Week of Jan 5, 2026PackagingFireworks Unifies Cached Pricing at 50% Across All Text and Vision Models
- Jan 10, 2026Price increaseFireworks Pricing Restructured: DeepSeek R1 0528 Sees 55% Input and 32.5% Output Price Reduction
Across Marketing & Content in the same window
All Marketing & Content changesQuestions about this change
Did Fireworks raise prices in March 2026?
Yes. Pulse captured Fireworks' pricing page on Dec 29, 2025 and again on Mar 30, 2026 and recorded 1 price increase: On-Demand: $4.00/hr → $6.00/hr (+50%).
What changed on Fireworks' pricing page between Dec 29, 2025 and Mar 30, 2026?
STT feature removed: Streaming ASR v1/v2; DeepSeek R1 0528 ($1.35 input/$0.68 cached/$5.4 output), Qwen3 235B Family ($0.22 input/$0.88 output), and Qwen3 Coder 480B ($0.45 input/$0.23 cached/$1.80 output) removed from serverless pricing table; On-Demand price: $4.00/hr → $6.00/hr (+50%). Pulse compared the captures it took on Dec 29, 2025 and Mar 30, 2026; the before/after table above lists each item as written on the page.
How often does Fireworks change its pricing?
This quarterly roll-up includes 4 changes Pulse first recorded in shorter periods and also records changes none of them did (price increase), so Pulse counts it as a change of its own; Pulse recorded 1 other pricing-page change for Fireworks in the 90 days before this one; the previous change was "Fireworks Pricing Restructured: DeepSeek R1 0528 Sees 55% Input and 32.5% Output Price Reduction" on Jan 10, 2026; Fireworks has been tracked since 2024-11-17.
