# OpenAI launches GPT-6 Sol and Luna at half the price the same week rivals ship too > Source: > Published: 2026-09-22 23:17:08+00:00 *OpenAI just cut its own API prices roughly in half with two new GPT-6 models, landing in the same 48 hours as fresh releases from Anthropic and xAI.* On September 22, OpenAI shipped two new models: GPT-6 Sol and GPT-6 Luna. Sol now costs $2 per million input tokens and $10 per million output tokens, down from $4 and $20 for GPT-5.6 Sol. Luna dropped to $0.10 and $0.50 per million tokens, from $0.20 and $1.20. A 50% cut, across the board. That's according to OpenAI's own announcement, confirmed by TechCrunch's coverage of the launch. The two models aren't interchangeable. Sol is built for complex coding and agentic work: the kind of multi-step task where a model has to plan, use tools, and check its own output. Luna is the workhorse. Cheap, fast, built for the high-volume jobs where you're running the same prompt a million times a day and every fraction of a cent matters. Both are already live in ChatGPT, Codex, and the API, for Plus, Pro, Business, Enterprise, and Edu users, with free-tier users getting Luna through the desktop app. ## The Rivals Didn't Wait Here's the part that makes the timing interesting: OpenAI didn't launch into empty air. Anthropic released Opus 5.5 on the same Tuesday, cutting output pricing to $20 per million tokens from $25. Not a small cut. It also claims real-world savings closer to 40%, since the model reportedly needs fewer tokens to finish the same work. According to TechCrunch, one early tester used Opus 5.5 to complete a 680,000-line code migration in under a day - a job that would normally eat an engineering team's whole week. Anthropic also says the model was tested externally by METR and Frontier Design before release, and that it's the strongest performer yet on the company's own automated safety audit. xAI moved first. Grok 4.7 shipped Monday, a day ahead of both. Unlike its rivals, xAI held pricing flat at $2 per million input tokens and $6 per million output tokens, the same as Grok 4.6. The model runs on 2.1 trillion parameters with extra training pulled from SpaceX engineering data, and it's already live in Cursor, GitHub Copilot, and the xAI API. On Artificial Analysis's benchmarks, though, Grok 4.7 landed just behind Claude's Fable 5.1 on both the AA-Briefcase and GDPval leaderboards. Decrypt summed it up bluntly: bigger, but late to the frontier party. [Four AI Labs Released Major Models in One Week and Buyers Can't Keep Up](https://startupfortune.com/four-ai-labs-released-major-models-in-one-week-and-buyers-cant-keep-up/) CNBC reports enterprise IT buyers are hitting 'model fatigue' after Anthropic, Meta, Google and OpenAI all released major new models within the same week in early September. Runpod CEO Zhen Lu says teams are burning outsized time just to avoid falling behind, while recent unauthorized-access incidents at Anthropic, Meta and OpenAI, all traced to... - [major AI model releases September 2026 enterprise buyers](https://startupfortune.com/four-ai-labs-released-major-models-in-one-week-and-buyers-cant-keep-up/) - [how to compare new AI models for enterprises](https://startupfortune.com/four-ai-labs-released-major-models-in-one-week-and-buyers-cant-keep-up/) ## What It Means For Builders If you're a founder deciding what to build on, this isn't background noise. Cost per token isn't a rounding error once you're running an agent that calls a model dozens of times to complete one task. A 50% price cut on Sol changes that math. And if Anthropic's 40% real-world savings claim holds, it changes the math for Opus too. You don't pick a frontier model because it topped a leaderboard once. You pick the cheapest one that clears your accuracy bar. That bar just moved for everyone at once. OpenAI is also leaning hard on an accuracy pitch it hasn't fully proven yet. The company says Sol and Luna make fewer factual and reasoning mistakes than the GPT-5.6 line, alongside gains in coding, computer use, and professional tasks. Those are OpenAI's own claims, drawn from its release notes. Independent benchmarks comparing Sol and Luna directly against Opus 5.5 and Grok 4.7 haven't landed yet. Anthropic, by contrast, put its safety and performance claims in front of outside evaluators before shipping. Not OpenAI. That's a real difference in how much you should trust the numbers on day one. None of this happened by accident. OpenAI has said it wants a model at every price point ahead of its DevDay event on September 29. Sam Altman has been explicit that the company sees this as a battle for who owns the default choice for builders, not just who tops a benchmark chart once. Three labs shipped three models in three days. Frankly, that's not a coincidence. It's a pricing war. And the ones who benefit most are the founders deciding which API key to put in production this week. **Also read:** [Snorkel AI Triples Its Valuation To $3.5 Billion On Surging AI Data Demand](https://startupfortune.com/snorkel-ai-triples-its-valuation-to-35-billion-on-surging-ai-data-demand/) • [Micron stock tops $1,000 for the first time as the AI memory boom rages on](https://startupfortune.com/micron-stock-tops-1000-for-the-first-time-as-the-ai-memory-boom-rages-on/) • [Anthropic Calls Opus 5.5 Pacing the Frontier While It Tops the Benchmarks](https://startupfortune.com/anthropic-calls-opus-55-pacing-the-frontier-while-it-tops-the-benchmarks/) *This article is posted in [AI News](https://startupfortune.com/category/ai/), check it out for more related stories.* ## Join the discussion [Open in the community →](https://startupfortune.com/community/) Almost there. Sign in and your reply posts straight away. [Anthropic Open-Sources Claude Commerce Agents Blueprint Ahead of Black Friday](https://startupfortune.com/anthropic-open-sources-claude-commerce-agents-blueprint-ahead-of-black-friday/) Anthropic has open-sourced Claude Commerce Agents, a free GitHub blueprint for building AI shopping and merchant agents, claiming pilot retailers saw carts grow up to 35% and purchases complete 60% more often. The release lands three months before Black Friday, just as OpenAI has pulled back from running checkout directly inside ChatGPT after weak... - [how to build AI shopping agents with Claude](https://startupfortune.com/anthropic-open-sources-claude-commerce-agents-blueprint-ahead-of-black-friday/) - [commerce agent blueprint for increasing customer cart conversions](https://startupfortune.com/anthropic-open-sources-claude-commerce-agents-blueprint-ahead-of-black-friday/)