Morning Edition · Friday, July 10, 2026
Tech
OpenAI Ships GPT-5.6 Family to the Public and Claims the Top Coding Benchmark
Sol Ultra scores 91.9 percent on Terminal-Bench 2.1 by spawning subagents, but its margin over Anthropic's Mythos 5 is smaller than the variation between two runs of the same model.
Tech
OpenAI Launches ChatGPT Work, an Agent That Runs Multi-Hour Tasks Across Apps
The product combines Codex into a single desktop app and adds a plugin directory spanning Google Drive, Slack, Salesforce, and GitHub.
Geopolitics
US Commerce Department Clears GPT-5.6 for Full Public Release
The model's earlier preview reached only pre-approved organizations under government-imposed access limits. That treated model access itself as a matter of export policy.
Tech
Developers Push China's Open-Weight GLM 5.2 Onto Modest Consumer Hardware
The MIT-licensed 753-billion-parameter model scores 62.1 on SWE-bench Pro and 81.0 on Terminal-Bench 2.1, and a community project now targets running it on slow local machines.
Tech
Ollama Raises $65 Million as Open-Model Tooling Reaches 8.9 Million Developers
The Series B, led by Theory Ventures, brings the local-inference platform's total funding to $88 million.
Tech
Meta's Brain2Qwerty v2 Decodes Typed Sentences From Brain Scans at 61 Percent Word Accuracy
The non-invasive pipeline reads magnetoencephalography signals, raising word accuracy from about 8 percent for prior non-invasive approaches, but it relies on a room-sized shielded scanner.
Tech
Survey Charts LLM Theorem Provers Moving From Solving Problems to Doing Research Math
The paper argues language-model provers are shifting from formal proofs of well-defined problems toward open research-frontier mathematics.
Tech
Theory Paper Asks When Reflection-Driven Reasoning Actually Helps LLMs
Using a sampling-complexity model of generate-critique-revise loops, the analysis characterizes the conditions under which in-context search outperforms plain sampling.
Tech
DeepSearch-World Trains Tool-Use Agents to Improve From Their Own Experience
The method uses self-distillation inside a verifiable environment to escape the reliance on fixed teacher trajectories and sparse reinforcement learning (RL) rewards.
Tech
AgentLens Argues Code-Agent Benchmarks Should Grade the Whole Trajectory, Not a Pass/Fail Bit
The production-assessed benchmark reviews the full run of interactive coding agents, targeting the gap between "task passed" and how the agent actually behaved.
World
EU Parliament's Effort to Kill 'Chat Control' Message Scanning Falls Short
A motion to scrap the interim rule drew 314 votes against 276 to keep it, but rejecting the Council's position required 360, so the scanning regime survives.
Tech
ByteDance Releases Seedream 5.0 Pro, an Image Model That Outputs Editable Layers
The professional model can split a finished poster into more than ten independent layers, inpainting the background behind removed subjects.