OpenAI Updates GPT-5.6 Sol for ChatGPT While Leaving API and Enterprise Versions Unchanged

OpenAI has upgraded the GPT-5.6 Sol model exclusively within the consumer ChatGPT interface, while keeping the implementations powering Codex and ChatGPT Work untouched. This marks a shift toward tailoring model variants for their specific deployment contexts.

Consumer-only optimization#

The company announced the update on Thursday, positioning the change as part of a consumer-focused refinement strategy. According to OpenAI's statement, "Because this version of GPT-5.6 Sol is optimized for everyday chats, it will only be available in the Chat experience in ChatGPT." The enterprise and API implementations remain untouched, creating what amounts to a branching of the same model name across different product lines.

New interface controls for fine-tuning answers#

Along with the model update, ChatGPT Plus and Pro subscribers now have access to a slider control across web, mobile, and desktop that lets users adjust how deeply the AI should reason through responses. This replaces the prior approach of maintaining separate models for quick versus contemplative answers. API developers have always had equivalent options, but they must continue evaluating whether the improved accuracy justifies increased latency and cost.

Reduced factual errors in benchmarks#

OpenAI reports that Sol now commits factual mistakes less frequently. Internal evaluations on financial, medical, and legal questions showed that answers with at least one error dropped by 68% compared to GPT-5.5 Instant. Luna, the model destined to power free tier accounts, achieved roughly a 62% reduction in similar error scenarios.

However, OpenAI did not release the test prompts or methodological details needed for independent verification. The comparison baseline was GPT-5.5 Instant rather than the previous version of Sol running in ChatGPT, making it unclear how much the current iteration improves over its immediate predecessor.

Monitoring systems remain in place#

OpenAI's System Card for GPT-5.6 discloses that both Sol and the Terra model are paired with classifiers that inspect generated answers in real time. When these systems flag potential issues, they trigger secondary verification before the response is delivered. OpenAI calibrates these classifiers independently for each model variant.