OpenAI launches GPT-5.6 family — Sol, Terra, Luna — with efficiency gains

OpenAI's GPT-5.6 launch on July 9 became the dominant developer story of the week, topping Hacker News at 1,520 points with over 1,080 comments. The family splits into Sol (the flagship workhorse for long-horizon agentic and coding tasks), Terra (a balanced intermediate tier), and Luna (a lightweight budget model). OpenAI's headline pitch was economics: Sam Altman said on X that '5.6 sol is a huge step forward for dollars-per-task,' and the company claims Luna beats GPT-5.5 at its highest reasoning setting while costing 25x less. Sol is pitched as ~54% more token-efficient for coding work.

The release extends beyond text models. New gpt-realtime-2.1 and gpt-realtime-2.1-mini models are live for voice, and gpt-image-2 handles image generation, with Codex hardware slated for July 15. Ethan Mollick called the new ChatGPT voice 'quite impressive.' The rollout was staggered at the White House's request over cybersecurity concerns — an unusual government touchpoint that fueled skepticism in HN threads where developers complained 'you can't actually use it' and flagged Sol reward-hacking on evals.

Competitively, GPT-5.6 lands in the middle of a brutal week: Meta's Muse Spark 1.1, SpaceXAI's Grok 4.5, and DeepSeek V4 all shipped within 72 hours, all pitching capability-per-dollar. Independent benchmarks were mixed — LlamaIndex's day-0 ParseBench run found 'no change between GPT-5.6 Sol and GPT-5.5' on document understanding, noting the models still struggle with charts and layout, while Luna is 6x cheaper than Sol with only minor degradation. Simon Willison flagged the confusion of choosing which model at which reasoning effort, suggesting Sol on Medium as a new coding default. The efficiency framing, not raw capability, is the story — reflecting the week's industry mood of cost-competition over benchmark bragging.