Together AI has launched a post-training service for open models such as Qwen, GLM and Kimi, supporting both supervised fine-tuning and reinforcement learning. Customers write their own training loop while Together handles the underlying infrastructure, and they keep ownership of the resulting models.
Vercel has added MiniMax's open-weight multimodal model M3.1 Flash to its AI Gateway, free of charge until October 22 when MiniMax itself serves the request. The model targets coding and agent work, offers a 1M-token context window, and MiniMax says it improves on earlier versions in agentic tasks, coding and cybersecurity.
Simon Willison has updated llm-echo, a debugging plugin for his LLM tool that provides a fake model which simply repeats the prompts it receives. Version 0.5 makes its simulated token counts account for tool calls, system prompts and reasoning tokens, which helps when testing changes ahead of a new LLM release.
The new OpenClaw release makes its update command finish and recover on its own across devices, file systems and platforms, instead of leaving a broken install. Interrupted turns, subagent work and deferred tasks now resume after a Gateway restart, so conversations are no longer lost.
NVIDIA has published its entire Run:ai GPU orchestration platform as open source under the Apache 2.0 license. The release includes the KAI Scheduler, GPU fractioning with CUDA 13.4, quota management and a local development environment based on Kind.
Google has released EmbeddingGemma 2, a compact open-weight multimodal embedding model based on Gemma 4 and licensed under Apache 2.0. It is designed to make semantic search and retrieval-augmented generation more efficient, including on devices.