At its first Forge conference on November 3, open-model inference platform Fireworks launched a batch of products. Ultra-fast, powered by NVIDIA Blackwell, is in public preview at up to 1,500 tokens per second on DeepSeek V4.1 Flash, cutting a 30-turn agent task from 1 minute to 8 seconds. Background Mode runs deferrable work at half price, alongside multi-day prompt caching and higher rate limits. Nexus Desktop brings open-model routing into the Codex app, Claude Desktop and OpenCode, with Pro plans from $8 per month. Fireworks also announced No Eval, serverless LoRA inference, the Fireworks Sovereign private preview and a 60+ partner Frontier Ecosystem.
Key Takeaways
- ✓Ultra-fast public preview: on NVIDIA Blackwell, up to 1,500 tokens per second on DeepSeek V4.1 Flash; in Fireworks' benchmark a 30-turn agent task fell from 1 minute to 8 seconds (Fireworks blog)
- ✓Cost options: Background Mode runs deferrable work at half price; multi-day prompt caching cuts the cost of resuming long context; higher serverless rate limits and Elastic Burst are generally available
- ✓Nexus Desktop launches: personalized routing across open and closed models inside the Codex app, Claude Desktop and OpenCode; Pro from $8 per month, Teams from $10 per seat per month
- ✓Training: No Eval (private preview) turns production traces into task evals; serverless LoRA inference (public preview) starts with GLM 5.3 Flash at per-token pricing with no idle charge; free training on Muse Glimmer 30B through November 30
- ✓Enterprise: Fireworks Sovereign (private preview) runs the inference engine in a customer's own cloud or data center; Fireworks says it already manages over 25,000 GPUs in customer environments; plus a 60+ partner Frontier Ecosystem
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and pain point
An agent often calls a model dozens of times per task, so latency piles up and cost climbs as more agents run at once. Enterprises also want to tune open models on their own data while meeting residency and audit rules. On November 3 Fireworks held its first Forge conference, and its official blog announced a batch of products aimed at these problems.
How it works
On inference, Ultra-fast runs on NVIDIA Blackwell GPUs for latency-sensitive work; Background Mode gives deferrable jobs such as eval data generation and overnight research a half-price runtime; multi-day prompt caching spares persistent agents from paying for their context again after a pause. Nexus routes across open and closed models, and Nexus Desktop now joins the Nexus CLI. Fireworks Sovereign brings the inference engine into a customer's own cloud, neocloud or data center, with inference and training inside that environment.
Benchmarks
All figures are Fireworks' own: Ultra-fast reaches up to 1,500 tokens per second on DeepSeek V4.1 Flash and cut a 30-turn agent task from 1 minute to 8 seconds; Fireworks' Ember-1 used about 40% fewer tokens than Kimi K3 at comparable quality in its evaluations. No independent reproduction has been published.
Getting started
Agent teams chasing latency can request the Ultra-fast public preview; batch jobs can move to Background Mode at half price. People using the Codex app, Claude Desktop or OpenCode can try Nexus Desktop, with Pro from $8 per month. Fine-tuning teams can train Muse Glimmer 30B for free with SFT, DPO and RL through November 30, then call the resulting LoRA on serverless at per-token pricing.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.