OpenAI launches its new family of models with GPT-5.6

OpenAI unveiled its newest household of models connected Thursday, introducing a caller acceptable of heavyweight programs into an progressively crowded tract of AI offerings.

GPT-5.6 comes successful 3 variants: Sol (considered its workhorse), Terra (a much intermediate option), and Luna (its fund affable option). These models grow what users tin bash crossed a assortment of fields — with the institution promising almighty capabilities successful endeavor work, coding, and adjacent technological research.

CEO Sam Altman has promised that his company’s newest models are orders of magnitude much businesslike and cost-effective than erstwhile versions, precocious telling CNBC that Sol is 54% much token businesslike erstwhile it comes to AI coding tasks.

Most notably, the institution calls 5.6 its “strongest cybersecurity exemplary yet, achieving frontier show with importantly less tokens.”

Indeed, overmuch hubbub has been made astir the model’s cyber capabilities, arsenic the Trump medication antecedently sought to restrict its rollout, ostensibly owed to fears of however the exemplary could beryllium misused. GPT-5.6 supports antiaircraft activities, including menace modeling, codification reappraisal and patching, and bluish teaming (simulating an onslaught connected your ain systems to find weaknesses earlier existent hackers do).

OpenAI besides released a caller instrumentality called ChatGPT Work, which — conscionable arsenic it sounds — is designed arsenic a workplace companion for endeavor teams, moving connected desktop, web, and mobile, that tin assistance with regular clerical tasks, similar drafting documents, spreadsheets, and presentations.

OpenAI’s recently announced household of models follows connected the heels of akin releases this week from competitors SpaceXAI and Meta.

However, GPT-5.6 and its attendant selling seems astir designed to instrumentality purpose astatine OpenAI’s superior opponent, Anthropic. Anthropic has managed to marque itself the likable underdog of the AI race, focusing fixedly connected endeavor customers and winning a increasing stock of enactment arsenic a result.

Not to beryllium outdone, OpenAI cites the Artificial Analysis Coding Agent Index, a notable benchmarking metric, to assertion that its latest household of models outshines Anthropic’s models astatine each turn.

OpenAI calls Sol its “best coding exemplary yet,” and has explicitly compared it to Anthropic’s recently released (and overmuch hyped) Fable. Using the Coding Agent Index, OpenAI claims that Sol “sets a caller authorities of the creation astatine 80, 2.8 points supra Fable 5, portion utilizing little than fractional the output tokens, taking little than fractional the time, and costing astir one-third less.”

It adds: “That vantage extends crossed the family: Terra performs conscionable supra Fable 5, portion Luna outperforms Opus 4.8.”

The institution says that 5.6 is present disposable crossed ChatGPT, Codex, and the OpenAI API. Availability per cardinal tokens is priced arsenic follows: Sol is $5 input / $30 output, Terra is $2.50 input / $15 output, and Luna is $1 input / $6 output.