# OpenAI halves Sol and Luna pricing after Astra _OpenAI launched GPT-6 Sol and Luna at 50% below GPT-5.6 promo pricing, adding faster, cheaper tiers beneath Astra with new caching gains._ **Published:** 2026-09-24 **Source:** https://www.startuphub.ai/ai-news/artificial-intelligence/2026/openai-halves-sol-and-luna-pricing-after-astra --- [OpenAI](https://openai.com/index/introducing-gpt-6-sol-and-luna/) introduced GPT-6 Sol and Luna to extend GPT-6 Astra to cheaper, faster tiers and cut API prices 50% versus GPT-5.6 promotional rates. Sol falls to $2 input and $10 output per million tokens, Luna to $0.10 input and $0.50 output. That pricing only works after Astra. Earlier this month [OpenAI](https://www.startuphub.ai/startups/openai) introduced GPT-6 Astra as its most intelligent and aligned model, then said work happens at different scales, rhythms and budgets. Sol and Luna were trained with similar methods as Astra and inherit its gains in professional work, factuality, coding, computer use and alignment. OpenAI says improvements in caching and inference let it serve the models more efficiently and pass savings to users, with 90% discounts on cached input-token reads still in place. The price ladder now has a clear anchor. Astra remains the top tier at $10 per million input tokens and $50 per million output tokens with a 1,050,000 token context window and up to 128,000 tokens of output. Sol sits below Astra on cost and above Luna on capability, Luna at the bottom for high-volume tasks. For buyers asking is GPT-6 API pricing real and is it legit, the numbers are real and published for API use, is it safe to adopt depends on workload fit rather than model risk. OpenAI frames the value as cost per task, not just tokens. On AutomationBench 1.0.6 across 47 tools, GPT-6 Sol at xhigh effort scores 33.2% at $0.27 per task, ahead of Claude Opus 5 at max effort at 26.9% and about 11.1 times its cost. On Agents’ Last Exam V1, Sol at max scores 56.4%, above Opus 5’s best at 60% lower cost. On [FrontierCode](https://www.startuphub.ai/ai-news/artificial-intelligence/2026/frontiercode-ai-coding-benchmark-goes-beyond-correctness) 1.1, which grades mergeability as well as correctness, Sol matches [Claude Fable 5.1](https://www.startuphub.ai/ai-news/artificial-intelligence/2026/claude-fable-5-1-hits-snowflake-in-private-preview) at xhigh at much lower cost. On DeepSWE 1.1, Sol at max hits 68.8%, within 1.1 points of Fable 5 at xhigh at about 80% lower cost, while Luna at max hits 66.6% at 93% less than Opus 5 at medium. On OSWorld 2.0 offline, Sol at xhigh reaches 60.5%, similar to Opus 5 at medium at about 80% lower cost. The caveats matter. OpenAI says evaluations were performed in its research environment or via API, which may differ from production [ChatGPT Work](https://www.startuphub.ai/ai-news/artificial-intelligence/2026/answer-engine-optimization-aeo-saves-atv-tour-days) due to system prompts and tools, competitor scores were taken from public reports, and the Fable 5.1 cost on AutomationBench omits Opus 5 fallbacks that occurred on about 40% of tasks. On factuality, Sol makes about half as many mistakes as its predecessor on an internal set of error-flagged conversations, approaching Astra-level reliability, but those conversations are not representative of typical use. On alignment, Sol and Luna improve over GPT-5.6 on deception, broken search and other probes, though tests deliberately create challenging conditions. Caching is the other lever for is GPT-6 API pricing worth it. Alongside lower token prices, OpenAI improved prompt caching to raise default hit rates and added a Prompt Caching Dashboard, diagnostics for missed cache opportunities, controls to adjust reasoning effort and tools without breaking cache, and explicit breakpoints to define cached prefixes. GitHub reports these changes cut the share of prompt tokens needing fresh processing by more than 50% across billions of requests to OpenAI models, which helps Copilot latency. What still has to happen is distribution. GPT-6 Sol and Luna are available in [ChatGPT Work](https://www.startuphub.ai/ai-news/artificial-intelligence/2026/frontiercode-ai-coding-benchmark-goes-beyond-correctness) and Codex starting today for Plus, Pro, Business, Enterprise and Edu users, with Luna also in the desktop app for Free and Go users, not yet in Chat, and as gpt-6-sol and gpt-6-luna in the API. OpenAI plans a gradual rollout in [ChatGPT](https://www.startuphub.ai/ai-news/artificial-intelligence/2026/chatgpt-images-bringing-ideas-to-life-in-any-size) throughout the day, so access may lag. For pros and cons, pros are lower price, higher cache hit rates and near-Astra behavior on everyday coding and business workflows; cons are that Astra is still required for the hardest projects and that headline scores and per-task costs depend on effort settings you must tune. The verdict for teams running agents is practical: if your workload is long-horizon coding or computer use at scale, Sol and Luna lower the price of iteration without giving up much accuracy. --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory.