Free ChatGPT goes unlimited; Sol upgrade cuts factual errors 68%
Luna becomes the default free model; elsewhere, an MIT SDK ships agents into Slack, Qwen3.8 Max tops the agent charts, and Fast mode handles 272K-token calls.
Copy markdown
The daily free-chat cap is gone
OpenAI is dropping the message limit on text chats for free and Go users and making GPT-5.6 Luna the default, replacing GPT-5.5. If your product’s free tier sits next to ChatGPT, the baseline just moved — unlimited chats and a new "Think" button roll out over the next week.
Sol’s error rate drops 68%
GPT-5.6 Sol lands as the new quick-task model for Plus and Pro with 68% fewer factual errors than GPT-5.5-Instant, plus a thinking slider to trade latency for depth. Sol is live today — worth re-testing any ChatGPT workflow that leans on fast factual answers.
Drop any agent into Slack with two npm installs
CopilotKit open-sourced the Channels SDK (MIT): it runs any AG-UI agent — LangGraph, CrewAI, Mastra, Pydantic AI, Google ADK — inside Slack and Teams, rendering native Block Kit and Adaptive Cards instead of plain text. Slack is GA, Teams is early access; a weekend away from shipping your agent where your team already works.
Qwen3.8 Max tops the agent charts
Alibaba’s open-weights Qwen3.8 Max (2.4T params) is now ranked the best overall model by the agentic index and scores 56 on Artificial Analysis’ intelligence index — ahead of every US lab except Anthropic and OpenAI. Picking a model for agentic or computer-use tasks? It’s a self-hostable contender worth benchmarking.
Fast mode now takes 272K-token prompts
OpenAI’s Fast mode now supports long-context requests above 272K tokens across GPT-5.6 Sol, Terra, and Luna, at up to 2.5x standard speed (for 2x the price). Whole-repo reasoning and long-transcript jobs no longer force you off the fast tier.