OpenAI launches GPT‑6 Sol and Luna

More affordable AI models for everyday tasks

OpenAI announced two new members of its GPT‑6 family, named Sol and Luna, to complement the earlier‑released GPT‑6 Astra.

Sol and Luna are built using the same training methods that powered Astra, but they are optimized for faster inference and lower operating costs.

The company says the new models sit on a “cost‑intelligence curve” that delivers strong capabilities while cutting expenses for users.

Pricing for Sol and Luna is cut in half compared with the promotional rates of the previous GPT‑5.6 series.

Specifically, Sol’s input price drops from $4 to $2 per million tokens and its output price from $20 to $10.

Luna’s input price falls from $0.20 to $0.10 and its output price from $1.20 to $0.50 per million tokens.

These reductions are described as a direct pass‑through of OpenAI’s infrastructure savings.

Benchmark results show higher efficiency

In the AutomationBench suite, which evaluates business workflows across 47 tools, GPT‑6 Sol at “xhigh” effort beats Claude Opus 5 at “max” effort while costing only 9 % of Opus 5 per task.

At the same high‑effort level, GPT‑6 Luna improves its predecessor by 5.4 percentage points and reduces cost per task by 58 %.

Additional data show Sol achieving a 33.2 % task‑score at $0.27 per task, outperforming GPT‑6 Astra at low effort (30.3 % score) which costs 3.9 times more.

Claude Opus 5’s max‑effort score of 26.9 % comes at 11.1 times the cost of Sol, and Claude Fable 5.1 with Opus 5 fallbacks scores 31.4 % at over 8.9 times Sol’s cost.

In the “Agents’ Last Exam” evaluation of long‑horizon professional workflows, Sol at max effort records a 56.4 % score, surpassing Claude Opus 5’s best result while delivering a 60 % cost reduction per task.

Factuality gains at lower price points

OpenAI’s internal factuality tests, based on de‑identified user‑flagged conversations, show Sol makes roughly half the mistakes of its GPT‑5.6 predecessor.

This error rate approaches the reliability of the premium Astra model despite Sol’s lower cost.

Luna also shows substantial factuality improvement; at higher effort levels it matches GPT‑5.6 Sol’s accuracy while costing about one‑hundredth as much.

The company frames these gains as evidence that advanced AI can be practical for a broader range of everyday applications.

OpenAI advises users to select Astra when they need the absolute best performance, but recommends Sol or Luna for most professional and automation tasks where cost efficiency matters.

By delivering high‑quality outputs at reduced prices, the new models aim to broaden access to sophisticated AI capabilities across industries.

Why This Matters: The 50 % price cut for Sol and Luna makes high‑performing AI usable for routine business workflows that previously could not justify the expense.