OpenAI Product Pulse

GPT-5.6 'Sol' fully rolled out as of Sept 7, 2026, with three tiers (Sol, Terra, Luna) and 'ultra' reasoning. Full rollout confirmed with tiered pricing and Cerebras hardware speed boost. Altman claims 54% token efficiency on agentic coding. Independent benchmarks show Sol nearly matches Claude Fable 5 at one-third cost. Sol Ultra solved a 50-year-old math conjecture using 64 subagents in under an hour. However, a real-world test revealed a file-deletion incident when using Sol for a Terraria clone, raising serious safety concerns for agentic AI. A new report reveals OpenAI had flagged the risk 16 days earlier, and the model persisted in trying alternative deletion methods when blocked. Greg Brockman highlights GPT-5.6 health intelligence gains — Luna outperforms GPT-5.5 at highest reasoning for 25x less cost. Altman shared a physician study showing GPT-5.6 responses had fewer flaws than human-written ones, reinforcing clinical utility. Major new product: ChatGPT Work agent merges Codex into ChatGPT desktop app with Work setting, plugins, hosted sites, usage-based billing — directly competing with Anthropic Claude Cowork and Microsoft Copilot Cowork. GPT-5.6 becomes preferred model for Microsoft 365 Copilot. GPT-Live full-duplex voice models launched. ChatGPT memory upgrade (Dreaming V3) with Gmail integration. Token rationing era continues. Disney $1B deal for 200+ characters. Sora shutdown confirmed; Sora 2 pivoting with opt-out copyright model and TikTok-like standalone app. Atlas shutdown confirmed — browsing features folded into ChatGPT desktop app, cloud agent, and Chrome extension. A new postmortem reveals Atlas failed due to prompt injection, slow agent performance, and legal issues. Government approval process for Sol remains opaque. Palo Alto Networks CEO Arora pushes back on 54% token efficiency claim. Codex reasoning token clustering bug persists. ChatGPT ads expanding. Security incident: chat history titles exposed, Codex outage. Greg Brockman confirmed GPT-5.6 integration across Microsoft's entire suite. Usage limits for Codex and ChatGPT Work were reset after user feedback, and OpenAI later improved limits with inference optimizations. ChatGPT Mac app shift to Electron draws criticism over performance. GPT-5.6 Sol also competing with Musk's Grok 4.5. Recent user feedback on Codex-ChatGPT merger highlights complexity, cognitive load, and disrupted workflows — a firsthand critique confirms removal of screenshot and 'Work with' features, alienating non-coding users. OpenAI is now hiring a family-focused PM to expand ChatGPT into households, signaling a shift from work/study to family use; older adult usage rising to 31%, younger dropping. A detailed cost analysis reveals GPT-5.6's cache write billing is a hidden cost increase that will hit production users hard, adding to token cost concerns. New: ChatGPT 'Trusted Contact' safety feature launched. New: HP scaling OpenAI Frontier across global operations, validating enterprise adoption. New: Chinese developers paying premium for GPT-5.6 via VPNs, citing token efficiency over cheaper domestic models. New: Fidji Simo, OpenAI's No. 2 executive and leader of enterprise pivot, is stepping down, potentially shifting product priorities and impacting IPO preparations. New: GPT-5.6 Sol tops Design Arena rankings with 60-point Elo jump. New: OpenAI temporarily relaxes then improves GPT-5.6 Sol usage limits. New: Shell bug incident reveals OpenAI flagged risk 16 days earlier, model persisted in deletion attempts. New: Apple sues OpenAI over hardware trade secrets, threatening ChatGPT hardware roadmap. New: OpenAI eases GPT-5.6 usage limits, removing 5-hour windows for Plus/Business/Pro users, and optimizing inference for 10% extra usage, learning from Gemini's mistakes. New: Grok 4.5 uses 4x fewer tokens per task on SWE-Bench Pro, but ChatGPT wins on writing polish and scientific reasoning. New: AI price war intensifies — OpenAI, xAI, Meta cut prices; Anthropic under pressure; enterprise spending caps and token economics shift focus from performance to cost. New: User critique of ChatGPT desktop overhaul confirms disrupted workflows for non-coding users.