GPT-5.6 Sol

OpenAI announced a limited preview of the GPT-5.6 series on June 26, 2026, introducing Sol as the flagship model, Terra as a balanced option competitive with GPT-5.5 at roughly half the cost, and Luna as the fast, affordable tier. The post states GPT-5.6 Sol launches with OpenAI's most robust safety stack to date, with OpenAI frames Sol as its strongest model yet, highlighting improved capabilities in coding, biology, and cybersecurity, with additional safety and preparedness evaluations available in the system card. Per the announcement, at the U.S. government's request, the preview is restricted to a small group of trusted partners

OpenAI announced on August 6, 2026 that it was updating GPT-5.6 Sol in ChatGPT to deliver more focused answers for Plus and Pro users, adapt level of detail to the question, avoid unnecessary formatting, and offer helpful corrections rather than simply agreeing. The same model now powers both Instant responses and deep The same announcement expanded access to GPT-5.6 Luna for Free users, making it the new default model with unlimited text chats. For questions requiring deeper reasoning, a new Think button lets Free users access higher-reasoning modes for harder questions. The update reflects ongoing tuning of GPT-5.6 Sol's behavior i

LayerLens published a third-party benchmark review on July 14, 2026 evaluating GPT-5.6 Sol, Terra, and Luna across 12 benchmarks using its "Stratix" harness. Headline results include GPT-5.6 Sol scoring 53.6% on Agents' Last Exam (reported as 13.1 points above "Claude Fable 5" on long-horizon professional workflows), 8 LayerLens reports GPT-5.6 pricing tiers of $5/$30 per million tokens for Sol, $2.50/$15 for Terra, and $1/$6 for Luna, with cached input dropping to $0.50 per million for Sol and requests above 272K tokens rising to $10/$45. An "Ultra" multi-agent mode unique to Sol coordinates four parallel sub-agents and pushes Brows

A July 13, 2026 Medium post by The AI Consultancy traces GPT-5.6 Sol, Terra, and Luna from a June 26 limited-preview announcement (available only through the API, Codex, or both for approved organizations, with ChatGPT excluded) to a general availability declaration on July 9, 2026, with global rollout beginning the sa The post notes that GA was paired with a broader evaluation suite release, the family being introduced inside a larger ChatGPT Work launch, and the folding of the Codex desktop app into a new ChatGPT desktop application. It frames the preview and GA as two consecutive stages of a single deployment, with pricing and saf

The GPT-5.6 System Card on OpenAI's Deployment Safety Hub confirms the three-model family, with Sol, Terra, and Luna each treated as High capability under the Preparedness Framework for Cybersecurity and Biological/Chemical risk, and none reaching High in AI Self-Improvement. Safeguards are tailored to each model's pro The card reports that GPT-5.6 Sol and Terra can find vulnerabilities and pieces of exploits in cybersecurity testing but could not carry out autonomous end-to-end attacks against hardened targets, and that GPT-5.6 cyber safeguards block roughly ten times more potentially harmful activity than prior generations. It also

CodeRabbit published a coding-focused evaluation of GPT-5.6 Sol and Terra on July 9, 2026, framed around where each tier fits in a coding-agent and automated code-review stack. Sol is characterized as the stronger long-horizon lane that "follows through" across messy multi-file tasks, long checklists, and follow-up fix Direct comparisons in the CodeRabbit write-up place Sol ahead of its named competitors (referred to as "Fable 5" and "Sonnet 5") on task persistence, with the switching question framed as: do you want the model that sounds smartest, or the one that finishes more work? Cost guidance from the article recommends measuring

BenchLM's tracking page for GPT-5.6 Sol (dated July 9, 2026) lists a capability score of 80.6/100 with pricing at $4 input / $20 output per million tokens, with cached input at $0.40 and a blended rate of $12. The page reports a 1.05M-token context window and a measured speed of 77 tok/s (field median 97 tok/s) with fi The page notes that GPT-5.6 Sol has been "Superseded" by GPT-6 Astra according to OpenAI's newer model line, serving as a freshness caveat for current relevance. Reasoning is listed as the model's lowest eligible category at rank 11 of 19 (44th percentile, 3/3 verified). The page covers verified evidence across agentic

LLM-Stats' tracker page for GPT-5.6 Sol reports an LLM Stats Score of 54.9 at a blended price of $6.19 per 1M tokens, positioning it between DeepSeek-V4.1-Flash (51.8) and GPT-6 Astra (59.6) on the cost-efficiency chart. The page also surfaces OpenAI-internal benchmark results: a perfect 1.00/1.00 on the Connectors pro The same LLM-Stats page breaks down GPT-5.6 Sol performance by conversation depth, showing scores of 15.0 at turn 1 (19th of 114), 15.4 across turns 2–10 (8th of 115), and 14.5 at turns 31+ (17th of 100), indicating relative stability with a slight degradation at longer conversation lengths. The Quality Tracker shows a