ChatGPT Gets a Reasoning Dial as OpenAI Folds Canvas and Atlas Into Core Chat
News Summary
OpenAI rolled out a sweeping set of changes to ChatGPT during the first week of August 2026, headlined by an upgraded GPT-5.6 Sol reasoning model for paid subscribers, unlimited text chats for Free and Go users, and a new manual slider that lets people control exactly how much computational effort the assistant spends on an answer. The update, announced August 6, 2026 (Eastern Time), arrives alongside a wave of related changes, including the retirement of the standalone Canvas writing and coding tool, the shutdown of the ChatGPT Atlas browser feature, and the phase-out of the original DALL·E GPT, signaling a broader push by OpenAI to consolidate ChatGPT's capabilities into a single, more flexible chat interface.
A Smarter GPT-5.6 Sol for Paid Users
The centerpiece of the update is an improved version of GPT-5.6 Sol, which now powers both the fast "Instant" mode and the deeper reasoning mode for Plus and Pro subscribers. OpenAI says the refreshed model produces noticeably more factual and focused responses, citing internal benchmarks showing 68% fewer factual errors compared to the earlier GPT-5.5 Instant model. The company also said it tuned the model's writing style to feel more natural in everyday conversation, cutting down on the overly long, bullet-heavy answers that had drawn user complaints since GPT-5.5's release.
A Slider for Controlling How Hard ChatGPT Thinks
Alongside the model upgrade, OpenAI introduced a new reasoning-effort slider available on ChatGPT's web, mobile, and desktop apps. The control lets Plus and Pro users manually choose how much thought the model puts into a response, moving between Instant, Medium, High, Extra High, and Pro-level thinking. Keeping the slider low favors quick answers for everyday questions, while raising it prompts ChatGPT to reason more carefully through tasks like coding, planning, research, or decisions that benefit from deeper analysis. The feature gives users direct control over a trade-off that was previously handled automatically or through separate model selection.
Free Tier Gets Unlimited Text Chats
Free and Go tier users are also seeing a significant expansion of access. OpenAI is removing the cap on text-based conversations for these users, who will now be able to chat without limits using GPT-5.6 Luna, a lighter-weight model tailored for everyday use. Free users are also gaining access to a new "Think" button that can be applied to harder questions, giving them a taste of the deeper reasoning previously reserved for paying subscribers, though OpenAI noted the expanded access remains subject to standard abuse safeguards.
Canvas Retired in Favor of In-Chat Writing and Coding
As part of the same update cycle, OpenAI confirmed that the standalone Canvas tool, previously used for drafting documents and writing code in a dedicated side panel, will no longer be available within GPT-5.5 Instant or GPT-5.5 Thinking. In its place, writing and coding functionality is now built directly into chat responses through inline writing blocks and code blocks, reducing the need to switch between separate views for collaborative editing.
Older Tools and Models Being Phased Out
OpenAI is also winding down several earlier products as it streamlines ChatGPT's feature set. The Atlas browser-based agentic feature is scheduled to stop working on August 9, 2026, with its capabilities being folded into ChatGPT and OpenAI's Codex coding tool instead. The original DALL·E GPT, one of ChatGPT's earliest specialized tools, is set for retirement on August 30, 2026, with OpenAI encouraging users to save any images they want to keep beforehand; image generation and editing will continue through the newer ChatGPT Images feature. Separately, the OpenAI o3 reasoning model is being retired from ChatGPT on August 26, 2026, following a 90-day sunset period after its successors became the default choice for most users.
New Sign-In and Academic Access Features
OpenAI also expanded ChatGPT's presence beyond the core chat product. A new "Sign in with ChatGPT" beta lets users create or link accounts on select partner platforms, including Airtable, GitLab, HubSpot, Notion, Supabase, and Vercel, using their existing ChatGPT credentials. Separately, the company launched ChatGPT for Academic Researchers, offering eligible faculty and postdoctoral researchers twelve months of complimentary access to a dedicated workspace with business-grade data protections, Pro-level usage limits, and collaboration tools for teams of up to five members.
Why It Matters
Taken together, the August 2026 changes reflect OpenAI's effort to make ChatGPT feel simpler and more capable at the same time: collapsing separate tools like Canvas and Atlas into the core chat experience, giving users direct control over reasoning depth through the new slider, and extending stronger models to free users who previously faced tighter limits. For the millions of people worldwide who rely on ChatGPT for research, writing, coding, and everyday questions, the update lowers the barrier to accessing more capable AI assistance while giving power users finer-grained control over how the system allocates its reasoning effort to each task.