Ollama
According to the Ollama release note, the api/show response now reports each model’s supported thinking controls and its default choice. Clients can therefore read this capability from the API instead of assuming it separately for each model.
Ollama
According to Ollama's release notes, excessive memory growth during long responses with MLX has been fixed. The stated fix concerns generation using speculative decoding; it is not an announcement of a general speed increase across all models.
OpenClaw
The release includes Atomic Updates that check the next version before switching over, allowing for safer deployment processes.
NanoJev
NanoJev publishes code, weights and training data for a Qwen3-0.6B-based decision model. Its game demonstrations also depend on shared planning code.
Google's updated CC agent coordinates information shared by up to six people. The experiment is initially available to adults with personal Google accounts in the US.
Qwen
On 18 September Alibaba's Qwen team unveiled Qwen3.8-Omni-Flash, which processes text, images, audio and video in a single request. It ships with a 1M-token context, tool calling and a thinking mode, and cuts audio-input cost by 98% versus its predecessor.
OpenAI
On 17 September OpenAI introduced Astra for Law, a legal configuration of GPT-6 Astra. It pairs a search index across more than 230 million US case-law, statute and court-rule sources with a zero-data-retention Trusted Access program, aiming to cut the made-up citations that embarrass lawyers.
TypeSafe AI
TypeSafe AI came out of two years of stealth to unveil Jev, its first model. Unlike LLMs, this 'System One' model generates no text and returns type-safe decisions. It is named after the Jevons paradox, and its founder helped build the methods behind ChatGPT.
AWS
Amazon Bedrock AgentCore's new runtime reclaims memory that goes cold during a session and keeps cold start times consistent regardless of container size.
Huawei
Huawei says its next-generation Ascend 960DT AI chip will now be ready in the first quarter of 2027, two quarters ahead of the earlier plan.
The UN System Data Commons, built on Google's open-source platform, lets users query statistics from nearly 20 UN agencies in natural language and supports MCP.
Crusoe
Data center developer Crusoe closed a $3.9 billion Series F that values the company at $30.9 billion.