Deepgram vs MiniMax

Deepgram acquires Daily to bolster voice AI infrastructure

Deepgram, a leading provider of speech-to-text and voice AI models, has announced the acquisition of Daily, the company behind the popular open-source voice AI framework Pipecat. The financial terms of the deal were not disclosed, but Deepgram has committed to keeping Pipecat open source. Pipecat manages voice and video streams, integrating STT, LLMs, and TTS while reducing latency. Deepgram aims to leverage Pipecat to enhance its real-time voice and video infrastructure, allowing developers to build expressive voice agents with pluggable models. Additionally, Deepgram is launching the Real-Time Context Standard (RCS), an open protocol co-developed with the Pipecat team, to standardize metadata in voice AI interactions. Daily had raised over $62 million from investors like Tiger Global and has been part of Y Combinator, like Deepgram.

• Deepgram acquires Daily, maker of open-source voice AI framework Pipecat
• Deal amount undisclosed; Pipecat remains open source
• Daily had raised over $62 million from Tiger Global and others

Introducing MiniMax Code Mini App

MiniMax has launched Mini App inside its MiniMax Code desktop app for macOS and Windows. Mini Apps are long-running applications featuring a frontend and a Node.js backend that install as plugins, enabling persistent interaction with AI agents. The release includes the Mini App runtime, a /miniapp-creator chat tool, and an open-source community repository.

• Launched on October 22, 2026, for local sessions of the MiniMax Code desktop app across macOS and Windows.
• Mini Apps combine a frontend, a Node.js backend, and a miniapp.json manifest compatible with the MiniMax Plugin protocol.
• Includes /miniapp-creator, allowing users to generate and test working applications directly from chat prompts using automated queries and screenshot verification.

MiniMax M3.1 Flash: Redefining Agent Performance and Cost

Today, we officially roll out MiniMax M3.1 Flash. As a next-generation Flash model, M3.1 Flash delivers a generational leap in agentic tasks, coding, Computer Use and cybersecurity, while making online inference nearly three times faster and reducing usage costs by 70%.

AI News Roundup: Model Launches, Policy, and Infrastructure

This AI news digest from mid-October 2026 covers major model releases, policy updates, and infrastructure advancements. DeepSeek's V4.1 Flash 1017 is highlighted as an open-weight value leader, featuring a 552B-total/16B-active MoE architecture with a 1M-token context and MIT license. MiniMax launched M3.1 Flash, offering competitive pricing and high-speed performance. Inception introduced Mercury 3, a diffusion LLM achieving over 1,000 tokens per second. Meta unveiled Muse Wonder Full and Muse Glimmer, including a 30B open-weight agentic model. Policy developments include a potential Chinese export rule and a DHS ban on DeepSeek. Infrastructure updates feature vLLM v0.32.0. Anthropic's new Public Benefit Fund commits about 20% of equity to public good, and OpenAI hired Aleksander Madry for safety.

• DeepSeek V4.1 Flash 1017 is a 552B-total/16B-active MoE model with MIT license, scoring 45 on AA Intelligence Index.
• MiniMax M3.1 Flash is priced at $0.15/$0.60 per million input/output tokens, with weights available Oct 23.
• Inception releases Mercury 3, a diffusion LLM with 1,000+ tok/s, available on Baseten.