not much happened today

2026-05-11 · Source: AINews · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Advanced, extended

Summary

Thinking Machines previewed "interaction models," trained from scratch for full-duplex multimodal interaction, enabling models to concurrently listen, speak, watch, think, search, and react. This represents a shift beyond turn-based AI, emphasizing continuous-time awareness and visual proactivity. Concurrently, OpenAI launched the OpenAI Deployment Company, a majority-owned unit with 150 Forward Deployed Engineers from the Tomoro acquisition, backed by $4B in initial investment, to help enterprises deploy frontier models. OpenAI also introduced Daybreak, an initiative for defensive cyber operations combining GPT-5.5, Codex, and specialized access tiers like Trusted Access for Cyber. Agent control planes are maturing, with tools like aggit, Claude agents terminal, and Cursor in Microsoft Teams emerging. New benchmarks like Artificial Analysis's Coding Agent Index are measuring model+harness combinations, revealing significant variations in cost, token usage, and time per task. Skepticism around TurboQuant is increasing, while local/open models continue to improve rapidly, with Qwen 3.6 35B A3B and DeepSeek V4 Flash showing strong performance on consumer hardware.

Key takeaway

For CTOs and VPs of Engineering evaluating AI integration strategies, recognize that the shift to natively interactive, multimodal AI is underway, demanding continuous processing capabilities rather than turn-based systems. Your teams should also consider the increasing maturity of local/open models like Qwen 3.6 35B A3B, which offer compelling performance and cost advantages for agentic workloads on consumer hardware, potentially displacing some hosted solutions within 12-24 months. Prioritize robust deployment and security frameworks, as demonstrated by OpenAI's Daybreak, to manage the operational risks of advanced AI.

Key insights

Native, full-duplex multimodal interaction and robust enterprise deployment are key next frontiers for AI.

Principles

• AI interaction should be continuous, not turn-based.
• Enterprise AI deployment requires dedicated engineering support.
• Local models improve faster than hardware ceilings.

Method

Thinking Machines trains "interaction models" from scratch for real-time, concurrent multimodal processing. OpenAI deploys models via a dedicated engineering unit and offers security-focused distribution tiers.

In practice

• Use Qwen 3.6 35B A3B for long-context local code refactoring.
• Employ SGLang for efficient multimodal model stacks.
• Benchmark agent systems as model+harness combinations.

Topics

• Multimodal AI Interaction
• OpenAI Enterprise Solutions
• AI Agent Orchestration
• Local LLM Inference
• AI Cybersecurity

Code references

• nathanlgabriel/paper_code_mapping_assessment
• Fringe210/llama.cpp-deepseek-v4-flash-cuda
• antirez/llama.cpp-deepseek-v4-flash

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Scientist, Machine Learning Engineer, AI Engineer

Related on AIssential

• Towards AI That Can Actually Interact — Real-time, continuous interaction models represent a fundamental shift from turn-based AI, enhancing human-AI collaboration.
• [AINews] Thinking Machines' Native Interaction Models - TML-Interaction-Small 276B-A12B - advances SOTA Realtime Voice and kills standard VAD — Real-time, full-duplex multimodal interaction models are advancing human-AI collaboration beyond turn-based systems.
• not much happened today — AI development is rapidly converging on composable agentic systems, multimodal understanding, and efficient local inference.
• [NOTICE] This feed is no longer maintained — Native interaction models enable real-time, multimodal human-AI collaboration by integrating interactivity directly into the AI's core architecture.
• Mira Murati steps back into the spotlight, carefully — Thinking Machines Lab is developing "interaction models" for real-time, continuous human-AI communication beyond turn-based prompts.
• The Sequence Radar #849: Last Week in AI: OpenAI Ships Agents, xAI Eyes Cursor, DeepSeek and Kimi Advance — AI is transitioning from standalone models to integrated, operational systems that coordinate action within workflows.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AINews.