OpenAI launches GPT-6 Sol and Luna: 50% cheaper, improved across all capabilities

OpenAI expanded the GPT‑6 family with two new models: GPT‑6 Sol and GPT‑6 Luna, designed to deliver advanced capabilities at lower cost. Both models bring improvements from GPT‑6 Astra—the company's most intelligent model released earlier this month—to faster, more affordable tiers. The models are 50% cheaper than their GPT‑5.6 predecessors and are available starting today in ChatGPT Work and Codex, as well as via API.

Pricing and availability

The new models represent a significant cost reduction:

Both models are available today in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users. Free and Go users can access GPT‑6 Luna in the desktop app. Gradual rollout is planned throughout the day. The API endpoints are gpt-6-sol and gpt-6-luna.

Performance improvements across tasks

OpenAI trained Sol and Luna using similar methods as Astra, carrying forward improvements in professional work, factuality, coding, computer use, and alignment.

Professional work: On AutomationBench (a test of business workflows across 47 tools), GPT‑6 Sol at extended reasoning outperforms Claude Opus 5 at maximum effort at just 9% of Opus 5's cost per task. GPT‑6 Luna improves on GPT‑5.6 Luna by 5.4 percentage points at 58% lower cost per task.

Factuality: On OpenAI's internal factuality evaluation (based on real-world ChatGPT conversations where users flagged errors), GPT‑6 Sol makes about half as many mistakes as GPT‑5.6 Sol, approaching Astra-level reliability at much lower cost.

Coding: On FrontierCode (which evaluates code readiness to merge into real codebases), GPT‑6 Sol matches Claude Fable 5.1 at extended reasoning. On DeepSWE v1.1 (complex software-engineering tasks in real codebases), GPT‑6 Sol at maximum effort scores 68.8%—within 1.1 percentage points of Claude Fable 5 at extended reasoning—at approximately 80% lower cost per task.

Computer use: On OSWorld 2.0 offline (long-horizon computer-use workflows), GPT‑6 Sol at extended reasoning achieves 60.5%, similar to Claude Opus 5 at medium effort, at approximately 80% lower cost per task. GPT‑6 Luna at maximum effort exceeds GPT‑5.6 Sol at medium effort at one tenth the cost.

Communication improvements

OpenAI brought GPT‑6 Astra's improved collaboration style to Sol and Luna. The models now show "more clarity, less jargon, fewer odd turns of phrase, fewer low-value details, and slightly shorter answers overall without losing substance," especially in technical and coding conversations.

Improved caching for agents

Alongside lower token prices, OpenAI improved prompt caching for GPT‑6 to deliver higher cache hit rates, helping developers reuse context at a 90% discount on cached input-token reads. New tools let developers monitor caching performance via a dashboard, adjust reasoning effort and tool availability without breaking cache, and optimize which prompt prefixes get cached. GitHub reported that these improvements have reduced fresh-processing token share by more than 50% across billions of requests.

Alignment improvements

GPT‑6 Sol and Luna build on alignment work from Astra, showing improvements over GPT‑5.6 counterparts, including lower rates of misleading claims about coding work. OpenAI's alignment evaluations deliberately test challenging situations and do not measure typical-use failure rates.

Why it matters

This release signals OpenAI's focus on cost efficiency as a key lever for AI adoption. By cutting prices 50% while improving capabilities, Sol and Luna make advanced AI practical for more everyday applications at scale—a critical shift as developer budgets become tighter and AI workloads grow. The improvements in prompt caching and inference efficiency also reduce operational costs for teams building agents and long-context applications. For SaaS builders and enterprises, the combination of lower prices and stronger performance across benchmarks (especially coding and computer use) expands the economic feasibility of AI-powered features and agents.