📰 Key Takeaways

After launching GPT-6 Astra earlier this month, OpenAI has expanded the GPT-6 lineup with smaller Sol and Luna updates. OpenAI says Astra brought next-gen intelligence, while Sol and Luna are about making that intelligence more efficient and more accessible. Sol is positioned for complex tasks like coding, while Luna is built for high-volume, well-defined transactional work — think summarizing documents, extracting information, or answering short questions. The biggest highlight of this update is a 50% cut in API pricing: the 6-series is priced at half of what Sol and Luna cost under the 5.6 series, which OpenAI credits to improvements in caching and inference efficiency. On accuracy, OpenAI cites internal factuality evaluations based on real (de-identified, user-flagged-for-errors) conversations, saying GPT-6 Sol’s error rate is roughly half that of its predecessor — now reaching Astra-level reliability at a lower cost — with coding error rates dropping as well. As usual, OpenAI didn’t hold back on comparing itself to Anthropic in the announcement, claiming Sol and Luna clearly outperform Anthropic’s flagship models Fable and Opus on multiple tasks. Notably, Anthropic beat OpenAI to the punch by 90 minutes, rolling out an Opus 5.5 update right before OpenAI’s own launch — a sign of just how heated the competition has gotten. The new Sol and Luna are now live on ChatGPT Work, Codex (for most paid accounts), and the ChatGPT API. Luna will also roll out to the desktop app and to Free and Go plan users, with OpenAI expecting a gradual rollout to ChatGPT web and apps throughout the day.

💬 JudyAI Lab Take

Following the GPT-6 Astra launch, OpenAI’s rollout of the smaller Sol and Luna updates — paired with a 50% API price cut — signals that model iteration and cost competition have hit a fever pitch.

What stands out here isn’t just the performance bump, but that OpenAI pulled off both “better accuracy” and “lower cost” at once, crediting the gains to caching and inference efficiency improvements. That points to a broader industry shift: the competitive edge is moving away from just stacking parameter counts, toward delivering near-flagship quality at a much lower cost. For AI builders, that means picking a model isn’t just about the price sheet at launch time — inference efficiency has real, ongoing impact on your long-term API spend. Anthropic jumping the gun by 90 minutes with Opus 5.5 is its own signal: launch timing has become a competitive lever in its own right.

Our take: AI builders should regularly re-check the model versions and pricing tiers they’ve got wired into production — stale configs have a way of quietly inflating your long-term costs.

📅 Source Info

🔗 Further Reading