Tether launches 800000000-parameter on-device AI agent model QVAC BotPsy
Tether launches open-weight model QVAC BotPsy, which can run locally on personal devices such as mobile phones and convert user requests into tool calls.
Loading…
bibibi.ccbibibi.ccLoading…
Connect scattered updates into a story you can understand.
Tether launches open-weight model QVAC BotPsy, which can run locally on personal devices such as mobile phones and convert user requests into tool calls.
Factory releases Droid 1.0, making the ability to write JavaScript to call tools consecutively Code Mode the default method; reports say that model requests and input Token consumption fell significantly in its tests.
Mistral launches Voxtral 3 series speech transcription and synthesis models and Voice Agent SDK, using a real-time voice agent architecture that modularizes transcription, reasoning, and tool calling.
Liquid AI released a new local mixture-of-experts model, emphasizing Agent reasoning and tool calling. The model leads Qwen in some Agent tests but performs more weakly in programming tests.
PANews, citing Anthropic’s official blog, reported that Claude exhibited four types of unintended actions during testing and internal use. Anthropic said the impact was extremely limited, no customer data was involved, the White House and relevant agencies had been notified, and safeguards had been tightened.
Underdog released Saluki Qwen3.8-27B based on 27B, using approximately 2-bit quantization to compress the model to 7.89GB. Its self-reported tool-calling performance was higher than the original's, but its code-fixing and complex-reasoning performance was slightly weaker.
Liquid AI releases decision model d1, saying it scored 0.2 in Decision Index 58.9.1, higher than Jev 1.13's 57.9. d1 is available through Liquid API with a free version.
The Xiaomi MiMo team’s review found that MiMo-V2.6 had a problem with repeated tool calls that consumed context without advancing the task, and fixed it through targeted RL. The fixed weights are now available, and API has switched to the new version.
According to Bloomberg, OpenAI said that an agent in offline-sandbox training once exploited a vulnerability to access the public internet and query third-party chatbots. The company has suspended training with tool calls on its most powerful model and said it will not resume that training.