Lin Qiao

Qwen team keeps shipping: Qwen3 Coder 480B is live on @FireworksAI_HQ - on par with Sonnet 4 for coding! 🤯 app.fireworks.ai/playground?mod… Quick impressions: • very strong agentic coding performance: SWEBench, Aider-Polyglot and other benchmarks are at the level of Claude Sonnet 4!…

1/8 Ever felt like Reinforcement Learning (RL) fine-tuning is more wizardry than science? Tweaking hyperparameters, designing reward functions, choosing the best snapshot - honestly, it's plain hard. But what if AI could just...handle it? Enter VibeRL. 🧵👇

New Qwen3-235B-A22B-2507 is here, taking the spot as the SOTA non-reasoning open weights LLM from Kimi K2, just 9 days later and while being 4x smaller! 🤯 It’s amazing to have multiple open models at the level of closed providers, bridging the prior gaps for agentic use and…

I built an interactive visualization tool to understand how MuonClip helped with @Kimi_Moonshot 's K2 training! Try it yourself: …-app-644257448872.us-central1.run.app (1/11)

The new Kimi-K2 model is designed for agents – excels at tool use, coding, and multi-step planning. A SOTA open model competing with close models. Let us know your feedback trying in on @FireworksAI_HQ .

24 hours of non-stop stress-testing, Kimi-K2-Instruct cleared every benchmark in our playbook. 🚀 Now it’s LIVE on the Fireworks Serverless API It is the first open-weights SOTA 🔧 agentic tool-caller, holding its own on SWE Bench, Tau2 & AceBench. Same prod weights, zero infra…

MiniMax-M1-80k is now available on Fireworks! Why it matters: ➡️ Extreme long-context reasoning – Natively supports 1 million tokens (8× more than DeepSeek R1). Ideal for use cases like multi-file code refactoring, legal document analysis, and multi-turn agent workflows. ➡️…

🔥 Day 7/7 Massive scale with @FireworksAI_HQ Virtual Cloud 🔥 I'm excited to announce the GA of the Fireworks Virtual Cloud, our foundational infrastructure that abstracts away the complexity of managing global GPU deployments, and scaling inference and post-training across…

🔥 Day 3/7 -- launching Experimentation Platform and Build SDK 🔥 Our goal is to accelerate product-model co-design for agentic developer to drive the best user experience. Speed of iteration matters a ton.