OpenAI replaced its split Instant/Thinking model lineup with GPT-5.6 Sol for ChatGPT Plus and Pro, per @rohanpaul_ai. The change unifies behavior across reasoning levels, addressing a long-standing user complaint.
Key facts
- GPT-5.6 Sol powers Instant and Thinking for Plus/Pro
- Free/Go users get GPT-5.6 Luna with unlimited chats
- Factual errors down 68% vs GPT-5.5 Instant (internal eval)
- New slider adjusts reasoning effort, model stays Sol
OpenAI has consolidated its ChatGPT model architecture. According to @rohanpaul_ai, GPT-5.6 Sol now powers both Instant and Thinking modes for Plus and Pro subscribers. Previously, moving from Instant to higher reasoning effort could feel like switching models because the tone and behavior changed with it. The new slider adjusts how much thought ChatGPT puts into a response while keeping GPT-5.6 Sol underneath.
Free and Go users receive GPT-5.6 Luna as the default model, with unlimited text chats and a Think button for harder questions. This mirrors the tiered approach OpenAI has used since GPT-5's rollout, but the key shift is behavioral consistency across reasoning levels.
OpenAI claims its internal finance, medicine, and law evaluation found factual errors were 68% less common than with GPT-5.5 Instant. The company did not disclose the evaluation's sample size or methodology, so the figure should be treated with caution.
Key Takeaways
- OpenAI unified ChatGPT reasoning under GPT-5.6 Sol for Plus/Pro, with Luna for free tiers.
- Internal eval shows 68% fewer factual errors.
Why the unified model matters
The move addresses a known UX friction: users who toggled reasoning effort often saw personality shifts, making outputs feel unreliable. By keeping Sol underneath across effort levels, OpenAI is betting that consistency builds trust more than raw capability jumps. This also simplifies model routing for the company, potentially reducing inference complexity.
What this means for users
For Plus and Pro subscribers, the practical change is subtle but significant: the same model now handles quick queries and deep reasoning, so tone and style remain stable. Free users get a more capable default with Luna and a manual Think button, narrowing the gap between tiers. The 68% error reduction claim, if replicated in third-party benchmarks, would mark a meaningful reliability gain over GPT-5.5 Instant.
What to watch
Watch for OpenAI's next developer documentation update to confirm GPT-5.6 Sol's API availability and pricing. Also track third-party benchmark results, such as SWE-Bench or GPQA, to verify the 68% error reduction claim. If Luna shows up in API logs, expect a broader rollout to enterprise tiers.
