Fireworks AI

Fireworks AI demonstrates that GLM 5.1, when using Claude Opus 4.7 as a sparse advisor in the Legal Agent Benchmark, achieves 18/100 all-pass versus 14/100 for Opus alone at 39% of the cost.

Step 3.7 Flash is a 196B MoE model designed from the ground up for inference efficiency, using MFA and AFD techniques to reduce KV-cache usage to ~22% of DeepSeek, supporting agent, coding, and multimodal workflows, open-sourced under Apache 2.0 and available on Fireworks.

开放权重模型在实际应用中表现出显著成本优势，RampLabs 使用 Kimi K2.6 和 DeepSeek V4 Pro 发现高危漏洞，成本降低 5 倍。

Fireworks AI provides a view that the success of a product lies in its uniqueness and adaptability as an RL environment.

Fireworks AI tests show baseline models had 20% retry rates in browser agent tasks, while Kimi K2.5/GLM-5/MiniMax M2.5 achieved near-zero retries with stable latency, directly impacting production system costs/delays/reliability.

Fireworks AI states that model fine-tuning now requires only a CLI command, 10 minutes of GPU time, and a few cents of compute cost, with full ownership of the weights. While 2026's off-the-shelf open models are sufficient for deployment, they remain just a starting point.

Fireworks AI introduces Step 3.7 Flash: a 198B sparse MoE VLM designed for inference from the start, with a 196B language backbone and 1.8B vision encoder, achieving up to 400 token/s on real-world agent workloads.

Routing and post-training open-source models significantly improve AI system accuracy, speed, and cost-efficiency. Harvey and Fireworks AI demonstrated that a hybrid architecture using GLM 5.1 as the primary worker with selective frontier model routing achieves superior quality and lower costs in legal tasks, proving this approach is a viable alternative to pure frontier models.

Fireworks AI 每日处理 30 万亿 token，开放模型使用量持续增长，Factory AI 的开放模型使用量在过去一个月内增长了 3 倍。

The Cursor team achieved Composer 2.5 through reinforcement learning training rather than prompt engineering, with their large-scale RL program running inference on Fireworks, indicating that self-trained models will be the only way to maintain competitive moats after 2027.

Cursor AI launched Composer 2.5 on the Kimi K2.5 base model, achieving 85% performance gains from reinforcement learning, with Fireworks AI providing the RL infrastructure for scalable deployment.

Fireworks AI 扩展其训练平台，支持 Nemotron 3 Ultra 模型的微调和 DPO 训练。

Day 2 at #MSBuild focuses on moving beyond generic foundation models, emphasizing customization, inference performance, and scalable deployment, with a live demo and real-world case study by @chahvivi.

The gap between fine-tuning and production inference is where teams get stuck; Fireworks AI at MSBuild discusses customization trade-offs, serving infrastructure choices, and scaling cost/latency optimization.

Fireworks AI introduces a new training update with Gemma 4 Dense, supporting 256K context training with SFT, DPO, or RL.

Fireworks AI launches Gemma 4 Dense model, supporting custom fine-tuning, 256K context, and multiple training methods.

Fireworks AI announced the addition of GLM 5.1 LoRA RL functionality to its training platform, supporting SFT, DPO, and full RL.

Fireworks AI announces its Kimi K2.6 model now supports full-parameter reinforcement learning (Full-Param RL), allowing users to fully customize training on it to achieve the maximum data moat.

NVIDIA Nemotron 3 Ultra 模型已在 Fireworks AI 上线，适用于复杂自主代理的推理和编排。

Fireworks AI announces the upcoming release of Microsoft MAI models, offering enterprise-level customized reasoning model services.

Fireworks AI demonstrated an end-to-end workflow for high-performance inference directly on Foundry at MSBuild, emphasizing how unified infrastructure reduces latency, cost, and simplifies deployment.

The article points out that using third-party APIs may allow competitors to gain the same capabilities, emphasizing the importance of building a durable moat through fine-tuned models and tight feedback loops.

Most teams can choose frontier models, but few can deploy them at production scale without facing constraints.

Fireworks AI announces a free default model for LangSmith Fleet, lowering the barrier for developers.

文章讨论了微调在AI模型中的重要性，但信息密度低，缺乏具体技术细节和实用建议。

Fireworks AI 被 Redpoint 的 InfraRed 100 列表收录，该列表表彰构建下一代 AI 基础设施的公司。

Fireworks AI announces a new era of hackathons, showcasing rapid development with AI tools and platforms.

Fireworks AI announced its partnership with LangChain for Deep Agents Inference at Interrupt 2026.

Fireworks AI announces its demonstration of AI tools on Microsoft Foundry and invites developers and enterprises to experience it at booth F111 during the MSBuild conference.

Production AI systems place very different demands on infrastructure once workloads scale, but the tweet provides no technical details or mechanisms, only serving as event promotion.