Google Releases Gemma 4, Microsoft-OpenAI Partnership Enters Its Next Phase, and NVIDIA Unveils a Multimodal Model

Google DeepMind Releases Open Model "Gemma 4"

Google DeepMind has released Gemma 4, the latest version of its open model family. The company touts it as "the most capable open model byte for byte." The edge-oriented E2B and E4B models prioritize multimodal capabilities and low-latency processing, with a designed focus on practical on-device usefulness.

As competition among open models intensifies, Gemma 4 is positioned to compete directly with Meta's Llama series and Mistral.

Microsoft and OpenAI Move into the Next Phase of Their Partnership

Microsoft published an article titled "The next phase of the Microsoft-OpenAI partnership," announcing that the partnership between the two companies has entered a new stage. The company emphasizes that advances in AI are grounded in Microsoft's mission to "empower every person and every organization on the planet."

Combined with the AWS sales agreement between OpenAI and Microsoft announced the previous day, it has become clear that the relationship between the two companies is being restructured from an exclusive arrangement into a more flexible one.

Gemini 3.1 Flash TTS and Gemini Robotics-ER 1.6

Google DeepMind also announced Gemini 3.1 Flash TTS, a next-generation text-to-speech model. With improved controllability of expressiveness and better efficiency, it is expected to be applied to real-time use cases.

In addition, the company released Gemini Robotics-ER 1.6 for real-world robotics tasks. It aims to strengthen robots' ability not merely to follow instructions, but to understand their environment and respond autonomously.

NVIDIA Unveils Unified Multimodal Model "Nemotron 3 Nano Omni"

NVIDIA has released Nemotron 3 Nano Omni, which integrates vision, audio, and language into a single model. Conventional AI agent systems have used separate models for vision, audio, and language, with the drawback of losing context during conversion — this model makes it possible to process everything in one pass.

It is already available on Amazon SageMaker JumpStart, and deployment to edge devices is also envisioned.

Anthropic Reports an Update on Claude Code Quality

The Anthropic Engineering Blog has published the findings of its investigation into reports that Claude's response quality had declined for some users. The post explains in detail how infrastructure noise affects agentic coding evaluations, including quantification of that impact, as well as efforts to improve quality.

Alongside this, a number of technical initiatives were introduced, including safety improvements for permission bypasses in Claude Code's auto mode and an experiment in building a C compiler with multiple agents.

AWS Publishes a Guide for Moving from Text Agents to Voice Assistants

The AWS Machine Learning Blog introduced an approach for migrating text agents to voice assistants using Amazon Nova 2 Sonic. It explains concrete methods for upgrading conventional text-based conversational systems into voice assistants capable of natural conversation.

---