🔥New model drop - Kimi K2-0905 is now available on Fireworks!.The latest model from @kimi_moonshot delivers significant upgrades for intelligent coding and developer workflows, matching Sonnet 4 on SWE-bench, terminal bench, and SWE-dev benchmarks. Deploy via API or try it in

RT @cline: Live in Cline, Kimi-K2-0905 via the @FireworksAI_HQ provider.

New demo! Turn natural language into SQL with reinforcement fine-tuning (RFT) on Fireworks AI. ✅+30% accuracy over base model.✅ 20%+ more accurate than larger SOTA models. All done with a fine-tuned Qwen 7B model + synthetic data. No production data needed.If you're building.

Honored to share that Fireworks AI made the 2025 Forbes Cloud 100! ☁️. Among 20 AI native companies on the list, 80% are Fireworks AI's customers or partners, including @cursor_ai, @NotionHQ, @vercel, @perplexity_ai, @mercor_ai, @clay_gtm, @huggingface and many others.

🔥 In this fireside chat, @the_bunny_chen (Co-founder, @FireworksAI_HQ) sits with @Aishwarya_Sri0 (Head of AI Developer Relations) to unpack DeepSeek V3.1 - one of the most powerful open-weight LLMs available today. They dive into how DeepSeek V3.1 pushes the boundaries of.

Your AI benchmark might be lying to you!.We ran a test where an AI-generated image scored 93.3% on a standard benchmark. It checked all 15 boxes on the evaluation list. Technically perfect!. But when you look at the image, it’s clear something’s off. It doesn’t resemble the

How do you go from a blank project to your first fully tested AI agent?. In our latest blog, we share how we adapted Test-Driven Development (TDD) into the LLM era, using evals to define desired behaviors before writing code. By supercharging Claude Code with Model Context

Try it now on Fireworks AI →

Last week, we introduced DeepSeek V3.1 on Fireworks AI. It’s a meaningful step forward compared to the previous V3 version, designed with real-world applications in mind. Here’s what’s new:. ✨ Hybrid reasoning modes: Toggle between “thinking” (chain-of-thought) for deeper

We ❤️ our partnership with @huggingface - and we're now their number 1 Inference Provider . Dashboard: You can spin up models on Fireworks AI at Thank you @huggingface team and @ClemDelangue @Thom_Wolf @julien_c for partnering

Fine-tuning isn’t just a feature- it’s foundational to building differentiated, production-ready models. At Fireworks AI, Supervised Fine-Tuning V2 (SFT V2) has been the backbone for teams fine-tuning open models at scale, across agentic workflows, retrieval systems, and.

When you optimize for the wrong rewards, models learn the wrong behavior. In our Reward Hacking tutorial, we walk through how to design structured, multi-signal rewards to guide LLMs toward trustworthy, relevant outputs- while avoiding shortcuts that lead to poor generations.

Announcing DeepSeek V3.1 is now available for fine-tuning on Fireworks, both SFT & RFT! 🔥. Customize the powerful new hybrid reasoning and agent capabilities of V3.1 for your use case. Optimize for quality, latency, and cost with our advanced Quantization Aware Training, plus

RT @lqiao: 🚀 Deepseek V3.1 a big step in coding and agentic progress. This release focused significantly on multi-step tool usage and agent….

What if you could explore your AI eval results like an Excel pivot table, without writing a single line of code?. Meet Pivot View in Eval Protocol. A visual, interactive way to slice and dice your evals across models, prompts, runs, scores, parameters, and more. → Want to know.

RT @the_bunny_chen: LLM-as-a-judge is not a silver bullet. If you really care about model quality, you can’t skip hands-on review of your e….

RT @the_bunny_chen: LLM-as-a-judge is the hottest new thing in AI. It's also a trap. Out of the box, they are uncalibrated, buggy, and will….

@OpenAI . cont 🧵. → The initial GPT-OSS release was a milestone, an incredible week for open-source AI! 🇺🇸 However, its tool-calling (function calling) had critical bugs. This meant unreliable outputs and failed actions, a blocker for agentic use cases. → We identified and fixed a.

We re-ran the @OpenAI gpt-oss model benchmarks, & the results are surprising!. We believe OpenAI under-reported its tool-calling performance by 5-10%. 😁. Here’s why, and how we fixed it. 🧵.