GPT-6 Sol and Luna landed on September 22, 2026 — barely 90 minutes after Anthropic released Claude Opus 5.5, in what TechCrunch called a direct reflection of “the intense competition” between the two labs. If yesterday’s Opus 5.5 piece was Anthropic’s move, this is OpenAI’s answer: two new models, both roughly half the price of their predecessors, both aimed squarely at the same high-volume coding and agent workloads Indian developers are increasingly building on.
We already broke down what Claude Opus 5.5 changes for Indian developers. This is the other half of that story — and for anyone choosing between the two stacks this week, the numbers below matter more than the marketing.
What OpenAI Actually Announced
- GPT-6 Sol is built for complex coding and agentic workflows; GPT-6 Luna targets high-volume, lower-complexity tasks like summarization and extraction.
- Both were trained using methods similar to flagship GPT-6 Astra, bringing down some of its gains in coding, computer use and factual accuracy at a fraction of the cost.
- On OpenAI’s internal factuality test — built from real, de-identified ChatGPT conversations where users had flagged errors — the company says Sol makes “about half as many mistakes” as GPT-5.6 Sol, calling it “approaching Astra-level reliability” at much lower cost. That’s OpenAI’s own quoted language, not an independent finding.
- Neither model is available in ordinary ChatGPT Chat yet. They’re live in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users, with Luna also reaching Free and Go users through the desktop app.
- There is currently no GPT-6 Terra — OpenAI hasn’t said whether one is planned.
Pricing: The Real Numbers
| Model | Input (per 1M) | Output (per 1M) |
|---|---|---|
| GPT-6 Sol | $2 | $10 |
| GPT-5.6 Sol (previous) | $4 | $20 |
| GPT-6 Luna | $0.10 | $0.50 |
| GPT-5.6 Luna (previous) | $0.20 | $1.20 |
| Claude Opus 5.5 (for context) | $4 | $20 |
Cached input reads are discounted roughly 90% on both new models. Context window sits at 1.05 million tokens. One pricing myth worth killing here: a few outlets claimed GPT-6 Sol “now matches Claude Sonnet 5’s price.” That’s outdated — Sonnet 5’s $2/$10 was promotional pricing that expired August 31, 2026; its standard rate is $3/$15. At $2/$10, GPT-6 Sol actually undercuts current Sonnet 5 pricing, not matches it.
The India Angle: Claude vs GPT-6, By the Numbers
India is ChatGPT’s second-largest market globally by weekly active users, and Claude’s India user base grew roughly 5x in the first half of 2026. Both companies are fighting for the same fast-growing developer base — which means this price war translates directly into lower monthly bills for Indian teams, regardless of which stack they pick. For coding-heavy agent workflows, GPT-6 Sol at $2/$10 is now cheaper per token than Claude Opus 5.5’s $4/$20, though cost-per-task (not just cost-per-token) is the number that actually matters, and neither company has published a fair head-to-head on that yet.
For Indian IT services firms and startups running both models in production, the practical move right now is testing token efficiency on your own workload rather than trusting either company’s launch-day benchmark chart.
Bottom Line
GPT-6 Sol and Luna are a real, verified price cut — not a promotional gimmick, per OpenAI’s own confirmation to The New Stack that this is now default pricing. Combined with Anthropic’s Opus 5.5 cut the same day, Indian developers have more frontier-capable, lower-cost options than at any point this year. The smart move isn’t picking a side — it’s re-running your own cost math on both.
References: OpenAI — Official Announcement · TechCrunch · VentureBeat · The New Stack · OpenAI API Pricing Docs