modal

@muni_bio 和 @onepot_ai 于七月联合举办 Ebolathon，团队在一天内设计出 140 个靶向 Bundibugyo 埃博拉病毒进入细胞的小分子，经合成与测试后获得 18.6% 的初步命中率，部分流程由 Modal 提供算力支持。

Modal announced first-class support for post-training Reflection's Beam model and released Modal Dojo, an open-source library and skill bundle for production-grade LLM training (including RL) on Modal, with an example training Qwen3.5-4B on GSM8K.

Modal and Gray Area present 'Scaling Intelligence,' a 5-day San Francisco exhibition featuring 11 newly commissioned artworks exploring AI and computing at scale, including a conversation on AI as a creative instrument.

Nova reports that its Agent OS, built with Modal Sandboxes, pushed Claude Opus 4.8 to 83.9% on Terminal-Bench 2.1—edging out Fable 5's 83.8% without any fine-tuning, showing that agent orchestration can now rival frontier model capability.

Tweet links to NumberGoUp.ai alongside an announcement that Modal reached $1B in annualized run rate with new Series D funding, highlighting its cloud platform for training, inference, and agent sandboxes.

Following a $750M Series D, Modal CEO Erik Bernhardsson discusses how each shift in software changes both the AI applications that can be created and the infrastructure needed to build them.

Modal 发布 Spindle，一个开源、与 Tinker 兼容的训练服务器，可在 Modal 工作区中运行，让现有 Tinker 脚本无需改动即可迁移，并以 GPU 时间计费取代按 token 计费。

Modal 分享了为 VM Sandboxes 构建的新文件系统栈的工程挑战，其自定义惰性加载容器文件系统 ImageFS 基于分层分布式文件系统（本地 NVMe、区域与 P2P 缓存、全局 blob 存储），在保持亚秒级冷启动的同时大幅提升性能。

Proximal 分享了其基于 Modal Clusters 构建的自研后训练基础设施，实现按需多节点集群与 GPU 自动扩缩容，以应对突发推理负载并打造高质量训练数据。

Opus 5.5 wrote a path tracer from scratch that traces up to 2,048 light paths per pixel, rendering realistic light interactions in 4K using Modal, without any 3D engine or graphics libraries.

Modal announced several new capabilities at Modal Runtime, including multi-node clusters, VM Sandboxes, Sandbox Sidecars, and Sticky Sessions, expanding its platform for scalable AI and compute workloads.

Modal's Runtime conference kicked off this morning with keynotes from Bernhardsson and Scott Wu, with product announcements and talk recordings to follow. Stay tuned for more details.

Modal teases an upcoming announcement happening tomorrow, signaling a new runtime offering for its cloud platform.

A speaker announcement for Modal's Runtime event in San Francisco, featuring talks on AI x Science, inference, training, and agents from engineers at NVIDIA, Typesafe AI, Runway, and Achira AI.

Modal is promoting its upcoming Runtime event, featuring a guest appearance from NEO, the humanoid robot by 1X Technologies.

ToposBio announced Topos-2, a next-generation model generating the full range of protein shapes across the order-disorder spectrum, plus a preview of Topos-Bind, a generative binding model built and run on Modal, aiming to drug previously undruggable targets.

Gergely Orosz shares his enjoyment of a 'GPU glossary' mini-book by Modal, obtained during a visit to their NYC office.

Modal has announced that Runtime will be available or launched in one week.

A fireside chat is happening live with Modal's CEO and co-founder, hosted by LangChain.

Modal details how they optimized inference performance for trillion-parameter coding agents, achieving significant improvements in throughput and interactivity for their service.