OpenAI Releases GPT-5.6 Sol: Right here’s How It Stacks Up In opposition to Different AI Fashions – Decrypt

OpenAI’s GPT-5.6 Sol is now public. The corporate launched its new flagship mannequin to common customers immediately, launching alongside two smaller siblings, Terra and Luna, after the U.S. Division of Commerce stored a preview restricted to about 20 trusted companions for 2 weeks.

The naming technique is new for OpenAI. That is the primary time the corporate provides names to its fashions as a substitute of merely utilizing numbers. Sol, Terra, and Luna mark functionality tiers that may transfer on their very own cadence. Sol is the flagship, Terra the on a regular basis mannequin OpenAI says matches GPT-5.5 at half the value, and Luna the cheaper choice.

Pricing runs $5 and $30 per million enter and output tokens for Sol, dropping to $1 and $6 for Luna. (Tokens, for these not within the know, are the smallest unit of data a mannequin can deal with. And corporations sometimes worth their fashions on a per token foundation for API companies) Two new knobs ship with it: a max reasoning effort that lets Sol assume longer, and an extremely mode that farms work out to subagents.

For context, Anthropic expenses $10/$50 for Claude Fable 5, Google expenses $2/$12 for Gemini 3.1 Professional, xAI expenses $15/$75 for Grok 4.5.

On the Chinese language facet: DeepSeek expenses $1.74/$3.48 for V4 Professional, and Xiaomi expenses simply $1/$5 for MiMo v2.5 Professional—putting Sol between the premium U.S. frontier fashions and China’s low-cost challengers.

What the benchmarks present

On Terminal-Bench 2.1—a check of command-line workflows that reward planning, software use, and iteration, scored because the share of duties a mannequin completes—Sol in its extremely configuration hit 91.9%, with customary Sol at 88.8%.

That places each forward of Anthropic’s Claude Mythos 5 at 88.0%, Claude Fable 5 at 84.3%, and Claude Opus 4.8 at 78.9%. Google’s Gemini 3.1 Professional Preview trailed the chart at 70.7%.

OpenAI leaned hardest on cyber. On ExploitBench, which measures how properly a mannequin finds and weaponizes software program vulnerabilities, Sol matched the restricted Mythos Preview whereas spending roughly a 3rd of the tokens. OpenAI says Sol nonetheless does not cross the “Cyber Important” line in its personal danger framework.

The testers have already got opinions

Early entry was loud. Theo, a widely known developer, AI youtuber and CEO of the AI platform T3 Chat, known as Sol “world main in pc use” and mentioned it mounted the complaints he had with GPT-5.5.

Dan Shipper, whose workforce at Each examined it for a month, supplied the cleaner line: “GPT-5.6 is sort of a Porsche, Fable is sort of a warp drive.” His learn is that Sol is the every day driver and Fable the factor you attain for to cross the galaxy.

Researcher Daichi Konno, who obtained in early, mentioned Sol clearly beats GPT-5.5 and lands close to Fable 5, with Anthropic nonetheless forward in writing duties. His sharper observe: Sol’s safeguards did not journey on life-science questions, which he thinks may make it a default for biology work.

Leaks already level previous it. One roadmap-tracking account, “Synthwave” on X, claims GPT-5.6 is the final of the 5.x line, with GPT-6 constructed on a bigger base arriving inside a couple of month. The identical thread pegs Anthropic’s Fable 5.1 as shut and DeepSeek’s V4 common launch as imminent.

The timing is pointed. Sol arrives the identical week Anthropic’s Fable 5 drops out of subscription plans—after it returned globally on July 1, the mannequin moved to utilization credit solely as soon as its 50% weekly allowance expired on July 7.

It is crowded on the high. SpaceXAI shipped Grok 4.5 yesterday at a fraction of the value, which Elon Musk known as “roughly akin to Opus 4.7, however a lot sooner.” Meta launched Muse Spark 1.1 this morning, its first paid mannequin. Neither leads the pack, however each are frontier-class.

That leaves Google because the one large U.S. lab that hasn’t refreshed its flagship on this run. Its high mannequin, Gemini 3, has been out since November 2025—the oldest frontier launch nonetheless standing whereas OpenAI, Anthropic, xAI, and Meta all moved in the identical seven days.