Skip to content
NewsModel release

OpenAI releases GPT-6 Sol and Luna at half the price of their predecessors, calling rates permanent

· by Pondero Newsdesk

The short version

OpenAI launched GPT-6 Sol at $2/$10 and GPT-6 Luna at $0.10/$0.50 per million tokens on September 22, 2026, cutting predecessor prices by 50 percent or more and marking the cuts as permanent.

OpenAI releases GPT-6 Sol and Luna at half the price of their predecessors, calling rates permanent

Luna's output token price dropped from $1.20 to $0.50 on September 22, 2026, set as a permanent rate rather than a promotional window. That number moves frontier-class summarization and extraction budgets onto a cost tier that many teams were already paying for older, lower-capability models.

What

OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, 2026, expanding the GPT-6 family below the flagship Astra model, per the company's announcement. Sol is priced at $2 per million input tokens and $10 per million output tokens, down exactly 50 percent from GPT-5.6 Sol's $4/$20 rate. Luna is priced at $0.10 per million input tokens and $0.50 per million output tokens. The input price is a 50 percent cut; the output price fell from $1.20, a 58.3 percent reduction. OpenAI described the new pricing as permanent, confirmed by VentureBeat as not introductory or time-limited.

Sol targets complex coding, agentic workflows, and professional knowledge work. On AutomationBench 1.0.6, Sol at xhigh effort scored 33.2 percent at $0.27 per task; Claude Opus 5 at maximum effort scored 26.9 percent at more than eleven times that cost per task, per OpenAI's published benchmarks. On DeepSWE v1.1, which grades agents on real-codebase pull requests, Sol at maximum effort reached 68.8 percent, within 1.1 percentage points of Claude Fable 5 at xhigh effort, at approximately 80 percent lower cost.

Luna targets high-volume, lower-complexity tasks including summarization and extraction. On DeepSWE, Luna at maximum effort scored 66.6 percent, comparable to Claude Opus 5 and Claude Fable 5 at medium effort, at 93 percent and 96 percent lower cost per task respectively.

Both models are live in the OpenAI API as gpt-6-sol and gpt-6-luna and available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Education users. Luna is also accessible to Free and Go users in the desktop app. Alongside the models, OpenAI released improved prompt caching for GPT-6. GitHub reported the improvements cut the share of tokens requiring fresh processing by more than 50 percent across production workloads. Cached input reads carry a 90 percent discount from the headline token rate.

Rerun your API cost models against Luna and Sol

Luna at $0.50 per million output tokens is the number that changes the most decisions. Many batch summarization and classification pipelines run on older GPT-3.5-class models chosen for cost. Luna closes the capability gap while matching or undercutting those price points. Teams with fixed monthly API budgets should rerun their cost models against Luna before the next contract cycle.

Sol at $2/$10 resets the mid-tier agentic comparison against every provider that has not yet moved on price. The AutomationBench and DeepSWE figures, attributed to OpenAI's evaluation suite, put Sol within 1.1 percentage points of Claude Fable 5 on coding tasks at roughly 80 percent lower cost per task. Teams running Codex or agentic coding pipelines get a direct cost lever without switching providers.

The 90 percent caching discount compounds both price cuts for systems that reuse stable prompt context. Agentic workflows with long, fixed system prompts see effective per-token costs well below the advertised headline rates.

OpenAI launched GPT-6 Astra, the flagship model, earlier in September 2026 at $10/$50 per million tokens. Sol and Luna complete the three-tier family by covering the cost segments below Astra where most production API traffic actually runs.

What to watch next

Luna's $0.50 output rate will pressure providers whose mid-tier pricing sits between $0.50 and $1.50 and have not yet responded. Whether Anthropic adjusts rates on its current lineup is the next signal. The more immediate question for operators is whether Luna handles their specific task mix at acceptable accuracy before the next budget cycle forces the comparison.

Sources