GPT-5.6 Sol

On August 6, 2026, OpenAI published a ChatGPT product update describing post-launch tuning of GPT-5.6 Sol for Plus and Pro users, emphasizing more reliable factual responses, tighter formatting, and answers that adapt their level of detail to the question. The same model now powers both Instant responses and deeper rea The update also expanded GPT-5.6 Luna access for Free-tier users, making it the default model with unlimited text chats and adding a Think button to invoke higher reasoning on harder questions. OpenAI noted that GPT-5.6 Sol is now used by roughly 1 billion people weekly across quick questions, web searches, planning, r

LayerLens published a GPT-5.6 benchmark review on July 14, 2026, evaluating the Sol, Terra, and Luna tiers across 12 benchmarks using their Stratix harness. The page explicitly reports GPT-5.6 Sol's performance, including 53.6% on Agents' Last Exam (beating Claude Fable 5 by 13.1 points on long-horizon professional wor The review also documents Sol's 1.05M-token context window retaining 91.5% on MRCR 8-needle recall at 256K-512K, and describes an "Ultra" multi-agent mode available only on Sol that pushes BrowseComp to 92.2% at 4x token cost. Tier-positioning guidance is concrete: Terra for development loops, Sol for production gates,

The llm-stats.com tracking page for GPT-5.6 Sol reports a composite LLM Stats Score of 55.0 at a blended price of $6.19 per million tokens. The page tracks performance across conversation depths, showing stable quality with a +0.79σ Quality Tracker reading based on 374 votes. Sol is ranked above Gemma 4 E4B (13.8), GPT Benchmark detail visible in the excerpt includes a perfect 1.00 score on OpenAI's Connectors production benchmark (measuring reliable connector-based tool use in agentic workflows) and 0.97 on internal Capture-the-Flag Challenges at max reasoning effort under OpenAI's Preparedness Framework. Performance metrics are bro

CodeRabbit published an engineering review on July 9, 2026, examining where GPT-5.6 Sol and Terra fit in coding-agent stacks and code-review workflows. The post explicitly names both Sol (flagship) and Terra (lower-cost) variants and frames Sol's distinguishing trait as long-horizon follow-through: persistence across f The review reports CodeRabbit harness results showing Sol finds more review issues than prior OpenAI models, while positioning Sol as stronger on practical task completion rather than architectural taste (where Fable 5 is described as still stronger). It notes Sol and Sonnet 5 trade wins across comment quality and revi

OpenAI's official preview post dated June 26, 2026, announces a limited preview of the GPT-5.6 series with three tiers: Sol as the flagship, Terra as a balanced model for everyday work competitive with GPT-5.5 at roughly 2x lower cost, and Luna as the fast and affordable tier. The post explicitly names GPT-5.6 Sol as " The post explains that the limited preview was conducted with a small group of trusted partners whose participation was shared with the U.S. government, and that OpenAI plans broader availability in the coming weeks. It teases upcoming evaluation results covering improved agentic capabilities in coding, biology, and cy