We evaluated Nexus on 7 benchmarks and 1,222 tasks. It defines the Pareto frontier. Mix open and closed models. Get better intelligence for 4x lower cost than a single frontier model. The smartest way to extend your Claude, ChatGPT, or Codex plan, starting at $8/mo.

Excited to welcome Dave Wehner to Fireworks as our CFO! Dave brings extraordinary experience helping build and scale Meta through some of its most defining stages of growth. What excites me most is that he is eager to roll up his sleeves and build again. He sees AI as the defining technology shift of his lifetime and believes in our mission. Our conviction is simple: every company should own its specialized intelligence. The next generation of AI leaders will build specialized intelligence from their unique data, knowledge, and expertise. Making this happen requires more than powerful models. It takes co-designed training and inference, personalized model ensembles, and radically better economics. We have an enormous opportunity ahead. I'm thrilled to have Dave join the journey! Welcome to Fireworks, Dave. We're just getting started!

Fireworks Forge may be over, but we're still glowing. Thank you to everyone who joined us yesterday. The speakers, partners and builders who were there made our first conference one to remember. Session videos are coming in the next few days. And we're already thinking about what's next. What was your favorite part?

lots of speculation about fireworks in sf tonight not related to elections we at @FireworksAI_HQ had our inaugural Forge conference and a namesake display is the best way to wrap it huge thanks to our incredible speakers, videos will be out shortly at

What an exciting day at @FireworksAI_HQ Forge! All things from evals, post-training, and custom continual intelligence with some of the sharpest people building at the frontier. Super excited to be working with the Fireworks team to bring that frontier to our customers @neosigma_ai. Thanks for having me!

I caught @DHH in the hall at @FireworksAI_HQ Forge. He mentioned his daughter got bored waiting for AI while building video games. That reminded me of a demo I built that we didn’t get to show: I kept asking Ultra-Fast to make a game cooler and cooler while Opus was still building the first version. @DHH, does she know how to do a barrel roll? Pew pew pew. 👇

Start with open models. Ensemble them to lower costs. Train on what makes your business unique. Today at Forge, we’re announcing new capabilities to make that practical for our customers, from AI-natives to Fortune 500. Details + request access:

Reflection is building Beam, a 501B-parameter MoE it plans to release with open weights. Today on the Glow stage, @MishaLaskin, CEO of @reflection_ai, joins our Co-founder @the_bunny_chen for a fireside chat at Fireworks Forge. Watch live:

Joining the Glow stage, @noahrshinn, Founder of @instinct sits down for a fireside chat with our CTO and Co-founder @dzhulgakov at Fireworks Forge. They discuss what he's learned building agents, and where they're going next. Watch live:

A keynote packed with exciting announcements by @FireworksAI_HQ during its Forge event: - Fireworks Sovereign: The Fireworks stack in any environment you need it - Ember Next-gen: Back in September, Fireworks Research released Ember-1, a specialized model that delivers Kimi K3's quality with 40% fewer tokens. Built on Kimi K3, it learned to cut unnecessary reasoning while keeping the thinking that matters. Today, Fireworks announces its next-gen: based on US open models, it targets 30x more efficiency than Ember-1 - No-GPU: Pay-per-token training and inference. No need to worry about the underlying infrastructure - No-evals: Evals that build themselves from traces and product signals - Ultra-loop: Recursive self-improvement for inference. Every experiment feeds a shared knowledge base, so each loop starts smarter than the last - Ultra-fast: Models can reach 1,500–2,000 tok/s - Ultra-efficient: Persistent agents (e.g., personal assistants) and background agents (e.g., swarms) are the emerging use cases in the industry. To support those: - Multi-day prompt caching: Builds on the already industry-leading cache hit ratio (95%+) - Background-mode: 50%+ lower cost for agents that can afford to wait Incredible work, congratulations to the team! Source: Forge 2026

More AI spend doesn’t mean better results. Choosing the right model for the job is what can get you there. Nate Yohannes and Sardhendu Mishra are at @FireworksAI_HQ Forge today talking about how fine-tuning your models can deliver stronger results faster and at scale. $ZETA

@JensenHuang's two predictions for 2027 at @FireworksAI_HQ forge 1. security depends on two things: open models plus a contained runtime - "next year we’re going to realize this one piece of software, which is a secure runtime, completely revolutionized AI...not because of what it can do, how we can safely deploy it" - treat agents like a new hire: minimal rights, always monitored, kill switch. - pair with strong open models (and open shell) so security isn’t one central point of failure. - open models enable “distributed self defense” - companies aren’t blocked on capability, they’re blocked on trust - which matters even more when agents start touching hardware 2. "hardware in the loop" is the next big frontier - “the second big breakthrough for the industry” - telecom: millions of radios already deployed, which could get far more spectrally efficient once agents control them - drug discovery: instruments from Thermo Fisher and others plus robot arms from KUKA and ABB, connected into one agentic system - manufacturing: eventually an entire fab where an agent runs every instrument - "an entire fab is really going to become agentic” - the bottleneck moves from the model to integration - a big opening for whoever builds the connective layer

A company brain that only answers questions never gets better at making decisions. @Alehav1, GM and Head of Applied AI at @Moodys, will be speaking at Fireworks Forge about how Moody’s is training specialized models on the lessons from its own decisions. Don’t miss these conversations, register to watch the livestream here:

A strong base model is only the starting point. Junbiao Tang of @AIatMeta will speak at Fireworks Forge on how Meta post-trains models for Business Agent and ad creative, plus Muse Spark and Muse Glimmer on Fireworks. Watch the keynote live Nov 3:

AI spend is one of the fastest-growing budget lines, and most finance teams only see it after the money's gone. @eleonorepigment, Co-CEO of Pigment, speaks at Fireworks Forge next week on how to forecast and control it at Fireworks Forge. Watch the keynote live Nov 3:

Fireworks Forge is set to be one of the best events of the year. When was the last time you saw this kind of talent in the same room at the same time? Check out @sophiamyang's post below for more info and to register for the livestream: