AI & Automation Briefing - September 24, 2026
OpenAI Releases GPT-6 Sol and Luna with Half the API Cost and Fewer Errors
OpenAI has expanded its GPT-6 lineup with updated Sol and Luna models, priced at half the API cost of the previous 5.6 series. The company attributes the price reduction to improvements in caching and inference. Sol targets complex tasks like coding, while Luna is built for high-volume, structured work such as document summarization and information extraction. OpenAI reports that GPT-6 Sol makes roughly half as many factual errors as its predecessor, reaching accuracy levels comparable to the flagship GPT-6 Astra at a lower cost. The new models are available now in ChatGPT Work and Codex for most paid accounts and through the API. Luna is also accessible via the desktop app and to Free and Go users. The release came roughly 90 minutes after Anthropic launched a new version of Opus 5.5, continuing the rapid back-and-forth between the two companies.
New Memory Architecture Lets LLM Agents Adapt Stored Experience to Each Task at Retrieval Time
A research paper from Salesforce and Nanyang Technological University introduces JitMem, a memory system for LLM agents that stores raw task trajectories without compression and curates them only when a new task arrives. Most current agent memory systems summarize past experience immediately after task completion, before the next query is known, which permanently discards potentially useful information. JitMem's read-time curator sees both the retrieved traces and the incoming task, then generates a compact memory payload tailored to that specific need. Because the payload is consumed on the same task, training uses a direct success signal rather than a delayed one. Tested across three benchmarks, ALFWorld, WebShop, and tau-bench, JitMem outperformed the strongest write-time baselines by 16.2, 16.3, and 3.9 absolute success-rate points. A single trained curator also transferred across different executor models. For teams building agentic workflows, the architecture suggests that how and when context is retrieved and filtered can matter as much as what gets stored.
ChatGPT Mobile Gains Voice-Driven Workflow Capabilities
OpenAI is rolling out voice-based agentic features to the ChatGPT mobile app, letting users trigger workflows through voice commands. Plus and Pro subscribers can use the mobile Work tab to create documents, draft emails, summarize Slack messages, build sites, create presentations, and access a cloud browser. Free and Go users get access to plugins and connected apps. OpenAI is also adding richer text output to voice conversations and enabling seamless switching between text and voice, with the ability to start a task on mobile and continue it on desktop. The update mirrors capabilities OpenAI previously brought to the desktop app after launching its GPT-Live conversational model in July. OpenAI is keeping its chat and workspace interfaces separate, unlike Anthropic, which recently merged its Cowork and Chat interfaces.