OpenAI has officially launched its highly anticipated flagship artificial intelligence model, GPT-5.6 Sol, marking a pivotal moment in the rapidly evolving AI landscape. This release is accompanied by two distinct, lower-tier models, Terra and Luna, signaling a new strategic direction for the company's product segmentation. The public debut follows a two-week period during which the U.S. Department of Commerce restricted access to a select group of approximately 20 trusted partners, generating significant industry buzz.
A New Naming Convention and Tiered Approach
Departing from its traditional numerical nomenclature, OpenAI is introducing a named system for its models. Sol, Terra, and Luna are designed to represent different capability tiers, allowing for more flexible, independent development cadences. Sol stands as the premium, high-performance flagship. Terra is positioned as the everyday workhorse, reportedly matching the prowess of GPT-5.5 at a significantly reduced cost. Luna rounds out the trio as the more economical option.
Pricing and Advanced Capabilities
The pricing structure for Sol reflects its premium status, set at $30 per million input tokens and $50 per million output tokens. Luna offers a more budget-friendly alternative at $1 per million input tokens and $6 per million output tokens. For context, input and output tokens are the fundamental units of information processed by AI models, with pricing typically scaled on a per-token basis for API services. Sol also introduces two novel features: a "max reasoning effort" knob, allowing the model to dedicate more computational time to complex tasks, and an "ultra mode" that intelligently delegates work to subagents for enhanced efficiency and performance.
Market Positioning and Competitive Edge
Sol's pricing places it strategically within the high-end frontier models from U.S. developers while remaining more competitive than some. For comparison, Anthropic's Claude Fable 5 is priced higher at $10/$50 per million tokens, and xAI's Grok 4.5 is even higher at $15/$75. Google's Gemini 3.1 Pro is notably cheaper at $2/$12. On the other hand, Chinese challengers like DeepSeek's V4 Pro ($1.74/$3.48) and Xiaomi's MiMo v2.5 Pro ($1/$5) offer significantly lower costs, positioning Sol as a compelling mid-to-high range option globally.
Benchmark Dominance and Early Acclaim
Performance benchmarks reveal Sol's impressive capabilities. On Terminal-Bench 2.1, a rigorous test assessing command-line workflow, planning, and tool utilization, Sol in its ultra configuration achieved an outstanding 91.9% task completion rate. The standard Sol model also performed robustly at 88.8%. These figures surpass Anthropic's top models, including Claude Mythos 5 (88.0%), Claude Fable 5 (84.3%), and Claude Opus 4.8 (78.9%), with Google's Gemini 3.1 Pro Preview trailing at 70.7%.
Early access users have showered Sol with praise. Theo, a prominent AI developer and CEO of T3 Chat, hailed Sol as "world leading in computer use," emphasizing its remarkable determination and ability to fix issues present in GPT-5.5. Dan Shipper of Every encapsulated its essence, stating, "GPT-5.6 is like a Porsche, Fable is like a warp drive" – positioning Sol as the ideal daily driver for demanding knowledge work and coding. Researcher Daichi Konno noted Sol's clear superiority over GPT-5.5, placing it near Fable 5, while highlighting its unyielding performance in life-science questions, potentially making it a default for biology-related tasks.
Cybersecurity Prowess and Future Outlook
In cybersecurity, Sol demonstrated significant efficiency on ExploitBench, matching the performance of the restricted Mythos Preview in identifying and weaponizing software vulnerabilities while consuming roughly one-third of the tokens. OpenAI asserts that Sol remains below its internal "Cyber Critical" risk threshold.
The release of Sol intensifies an already fierce AI arms race. It arrives concurrently with Anthropic's Fable 5 transitioning to a usage-credit-only model and hot on the heels of xAI's Grok 4.5 and Meta's first paid offering, Muse Spark 1.1. Notably, Google's Gemini 3, released in November 2025, stands as the oldest flagship model among major U.S. labs without a recent refresh. Meanwhile, whispers of future innovations, including a potential GPT-6 within a month and updates to Fable and DeepSeek models, suggest the AI frontier is set for continuous, rapid expansion.