OpenAI Launches GPT-5.6 Family: Three Tiers Roll Out with Focus on Sol Cost-Efficiency

On July 9, OpenAI officially launched the GPT-5.6 family, centering around three distinct performance and price tiers: Sol, Terra, and Luna. This release is widely seen as an effort to clearly define the gradient between model capabilities and costs, directly impacting developer selection and cost structures.

Key Details and Cost-Efficiency

According to a summary by @docdavkitty, the three tiers are fully online. @sama highlighted the cost-efficiency of the main tier, Sol: in many tasks, GPT-5.6 Sol is priced at half of Fable's cost with approximately double the token efficiency. He also stated that OpenAI will continue to reduce delivery costs to a quarter of current levels. @haider1 and @JustinHalford echoed this view, noting that Sol achieves performance close to Fable at a lower cost, which might diminish the presence of Fable 5 if it remains restricted to the API.

Rollout and Early Feedback

The GPT-5.6 family has gradually rolled out across ChatGPT, Codex, and the API. @luisdans noted that the three new models also appeared in Microsoft Scout's model selector. Regarding evaluations, @soumitrashukla9 cited the Agent Arena leaderboard, showing that GPT-5.6 Sol ranked second in browser-based tasks. Furthermore, a post forwarded by @steipete mentioned that GPT-5.6 boosted Z.ai's throughput by 60%, reflecting the new model's significant impact on actual deployment efficiency. @thione also mentioned the simultaneous launch of ChatGPT Work and a new desktop app. However, @TheZvi cautioned that the early hype surrounding this release might be somewhat biased.

2026-07-13 ~ 2026-07-15 · 10 related posts

Full story(20 episodes)→

Primary sources

1 near-duplicate retellings: eyishazyer