OpenAI Launches GPT-5.6, New Model Family Focused on Coding

OpenAI released a new GPT-5.6 model family on July 9, which the company calls the biggest leap in coding capability since the launch of GPT-5. The family consists of three variants - Sol, Terra and Luna - and is available immediately in ChatGPT, Codex, and through the OpenAI API.

Flagship Sol is billed as OpenAI's strongest model yet for coding, cybersecurity and scientific work, including threat modeling, code review, vulnerability patching and blue-team tasks. Terra is the variant aimed at everyday business work, while Luna, the cheapest version, is meant for applications where the cost of a single query matters most.

Three variants, one core

All three models share the same architecture and skill set, differing mainly in scale and price. OpenAI emphasizes that even the cheapest variant, Luna, outperforms the earlier Opus 4.8 model on coding benchmarks, despite a much lower per-token price.

The company is positioning the 5.6 family as a response to growing competition in the agentic model segment, where business customers increasingly judge providers not just on the quality of a single answer but on the cost and speed of completing entire coding tasks handed off to an agent.

Results against the competition

According to the independent Coding Agent Index ranking, Sol scored 80.2 points, 2.8 points higher than the previous leader, Anthropic's Fable 5. OpenAI notes that Sol achieved this score while using less than half the output tokens of its competitor and in less than half the time Fable 5 needed, at a cost roughly one-third lower.

The model is 54 percent more token-efficient on agentic coding tasks - Sam Altman, CEO of OpenAI

Pricing and Microsoft integration

Pricing per million tokens is $5 for input and $30 for output for Sol, $2.50 and $15 for Terra, and $1 and $6 for Luna. That's roughly in line with the previous generation of models, despite noticeably higher performance on coding tasks.

Alongside the launch, OpenAI also announced that Sol will become the default model powering Microsoft 365 Copilot in Word, Excel, PowerPoint and the new team collaboration app Cowork. It's another step in the two companies' long-running partnership and a sign that Microsoft is, for now, betting on OpenAI's models even as it works on its own MAI models in parallel.

For Polish companies using Copilot or the OpenAI API directly, the biggest practical change will be the price-to-performance ratio. Lower token consumption on agentic tasks means, in practice, lower bills for the same coding or data-analysis jobs, even without any change to per-token rates themselves.

The GPT-5.6 launch comes amid an accelerating race among models specialized for developer work. A few days earlier, xAI unveiled Grok 4.5 as a cheaper rival for coding tasks, and Google is preparing to launch a heavily overhauled Gemini 3.5 Pro. By rolling out Sol, Terra and Luna at once, OpenAI is trying to cover the market's entire price range in a single move.

Sources: OpenAI (openai.com), TechCrunch (techcrunch.com), CNBC (cnbc.com), ITwiz (itwiz.pl)