Reported News
Fireworks Forge: Ultra-fast inference preview, half-price Background Mode and Nexus Desktop
Fireworks announced a batch of launches at its Forge conference on November 3, 2026, led by Ultra-fast inference in public preview, which it says reaches "speeds up to 1,500 tokens per second" on DeepSeek V4.1 Flash. It also added a half-price Background Mode, Nexus Desktop with Pro plans from $8 a month, and serverless serving for trained LoRA adapters.
- Changes reported
- 6
- listed below, each from a source we read
- We tested
- Not tested yet
- reported from sources; we did not run it ourselves
- Gaps we flag
- 3
- things we could not verify
- Sources
- 1
- 1 primary, linked below
What changed
| # | What the source says | Status |
|---|---|---|
| 01 | Ultra-fast inference on NVIDIA Blackwell GPUs is in public preview. Fireworks says it reaches up to 1,500 tokens per second on DeepSeek V4.1 Flash, and that a benchmarked 30-turn agent task fell from 1 minute to 8 seconds. | Reported |
| 02 | Background Mode is a cheaper runtime at half price for work that can wait, such as eval data generation or overnight research. Fireworks also announced multi-day prompt caching and higher serverless rate limits (now generally available) for agents that pause and resume. | Reported |
| 03 | Nexus Desktop joins the Nexus CLI and routes open and closed models inside the Codex app, Claude Desktop and OpenCode. Pro plans start at $8 per month and Teams at $10 per seat per month; Enterprise adds SSO, governance and shared budgets. | Reported |
| 04 | Specialized Model Inference (public preview) serves trained LoRA adapters through the serverless API with per-token pricing and no charge while idle, starting with GLM 5.3 Flash. Training on Muse Glimmer 30B is free through November 30. | Reported |
| 05 | No Eval (private preview) turns production traces into task-specific evaluations, and Fireworks Sovereign (private preview) runs its inference engine in a customer's own cloud or data center. | Reported |
| 06 | Model governance, spending controls and exportable audit logs are now generally available, with regional data controls and customer-managed encryption in preview. Elastic Burst, which lets usage exceed its normal ceiling when spare capacity exists, is also generally available. | Reported |
reported by the sourcemeasured or run by usnot verified
Primary source: Fireworks blog: Fireworks unveils new products at Forge. We are the fast, independent index; the vendor’s own page is the source of record for this release.
In the source’s words
“in a benchmarked 30-turn agent task, completion time fell from 1 minute to 8 seconds at ultra-fast”
“New Pro plans start at $8 per month, and Teams plans start at $10 per seat per month.”
“Fireworks already manages over 25,000 GPUs in customer environments.”
Early verdict
Most relevant to teams running multi-step agents on open models: Ultra-fast targets latency and Background Mode targets cost, but both are vendor-benchmarked and Ultra-fast is still a preview.
Basis: Based on the Fireworks announcement, which states the speed, half-price and preview status; we have not tested either mode.
What we have not verified
- We did not run Ultra-fast, Background Mode or Nexus; the 1,500 tokens per second and 1 minute to 8 seconds figures are Fireworks' own benchmarks.
- The post does not give per-token prices for Ultra-fast or say which models Background Mode covers beyond "eligible work".
- No Eval and Sovereign are private previews; we did not check who can get access or when they reach general availability.
This brief is Reported: written from the sources below, not from running the release ourselves.
Sources
- Fireworks blog: Fireworks unveils new products at Forgeprimary source · fetched 2026-11-03
Related
Spotted an error? See our corrections policy or email the address on the contact page. Markdown version for agents: /news/fireworks-forge-ultra-fast-nexus-desktop.md
Want these briefs in your Google Top Stories? Add Agentic Weekly as a preferred source on Google.
Newsletter
The Weekly Verdict. Free, every Monday.
A free email every Monday: the week's few big AI releases, then 5 short use cases for the new models, each with a prompt you can paste.
Free. Double opt-in. One click to unsubscribe.