GPT-6 Sol and Luna Land 90 Minutes After Claude Opus 5.5: Inside the September 22 AI Price War

TL;DR: On 22 September 2026 Anthropic released Claude Opus 5.5 at $4 per million input tokens and $20 per million output tokens, 40% below Opus 5. Roughly ninety minutes later OpenAI shipped GPT-6 Sol ($2 / $10) and GPT-6 Luna ($0.10 / $0.50), each half the price of the GPT-5.6 models they replace. Both companies are pushing frontier-class capability into a price band that was mid-tier a year ago, and the winners are developers who re-evaluate their model choices this month.

Three weeks after OpenAI introduced GPT-6 Astra as its most capable and most expensive model, the company has filled in the rest of the GPT-6 family. Anthropic, meanwhile, chose the same Tuesday morning to launch Claude Opus 5.5. TechCrunch's report on the OpenAI launch notes the timing precisely: Anthropic's announcement went out about 90 minutes before OpenAI's.

Here is what actually shipped, what it costs, and what it means if you are paying for tokens.

What OpenAI released: GPT-6 Sol and GPT-6 Luna

OpenAI's framing, quoted by TechCrunch, was that "GPT-6 Astra introduced a new generation of intelligence; these models extend its benefits by making that intelligence more efficient and accessible."

The two new models sit below Astra in the lineup:

Source: OpenAI's model pages for GPT-6 Sol, GPT-6 Luna and GPT-6 Astra, and the OpenAI API pricing page, all read on 25 September 2026.

Some details from those pages matter more than the headline prices:

• All three models share a 1,050,000-token context window, with a maximum of 922,000 input tokens and 128,000 output tokens per request. The long window is not free, though: any request with more than 272,000 input tokens is billed at 2x the input and cache rates and 1.5x the output rate for the whole request.
• Reasoning effort is configurable from none (Sol and Luna only) through low, medium, high, xhigh and max. On the Chat Completions API, function calling is only supported when reasoning_effort is none; the Responses API has no such restriction.
• Cached input is 10% of the input price. A repeated system prompt or retrieved document costs $0.20 per million tokens on Sol and one cent per million on Luna.

OpenAI told TechCrunch that Sol makes "about half as many mistakes" as its predecessor on the company's internal factuality evaluation, and that the API pricing is half that of the GPT-5.6 series. The published pricing page confirms the second claim: GPT-5.6 Sol lists at $4 / $20, exactly double GPT-6 Sol.

Availability at launch: Sol and Luna in ChatGPT for most paid accounts and in Codex, both in the API, and Luna for Free and Go users in the desktop app, with a gradual rollout to the web and mobile apps over the day.

What Anthropic released: Claude Opus 5.5

Anthropic's announcement leads with cost: Opus 5.5 costs "40% less" to run than Opus 5, cache reads cost 60% less, and output generation is more than 30% faster. The list price is $4 per million input tokens and $20 per million output tokens, with cache reads at $0.20 and cache writes at $5. A "fast mode" with higher throughput is priced at $8 / $40.

The benchmark table Anthropic published:

Anthropic also says the model performed better in alignment testing than Opus 5, is less vulnerable to prompt injection, and consumes fewer tokens on equivalent tasks, which compounds the price cut for agentic workloads that run for many turns.

Opus 5.5 is available on the Claude API as claude-opus-5-5, in Claude Code, and on AWS, Google Cloud and Microsoft Azure. Like every Claude model from 4.6 onwards, it has a 1,000,000-token context window at standard pricing, according to Anthropic's pricing documentation. Claude Sonnet 5.5 and Claude Haiku 5.5 "will follow in the coming weeks."

The same pricing page carries a quieter piece of news: Claude Sonnet 5's $2 / $10 launch price, which was supposed to rise to $3 / $15 on 1 September, is now permanent. The increase "will not occur."

The price war in one table

Put the two launches next to the rest of the market and the pattern is obvious. Prices below are list prices per million tokens as published on 25 September 2026; see our LLM API pricing comparison for the full table with sources.

Two things stand out. First, the "workhorse" tier has converged on roughly $2 to $4 for input and $10 to $20 for output, with OpenAI now undercutting Anthropic's Opus line by half. Second, GPT-6 Luna is priced against open-weight hosted models such as DeepSeek V4.1-Flash rather than against anything in OpenAI's own catalogue, which is a direct response to how much of the high-volume market has moved to cheap open models.

What to do about it

Re-run your evals this week. If you standardised on GPT-5.6 Sol or Claude Opus 5 earlier this year, the direct replacements are cheaper and, by their makers' accounts, more accurate. Swap the model string in a staging environment, run the same evaluation set, and compare cost per successful task rather than cost per token. Our guide to evaluating retrieval quality covers how to build that evaluation set if you do not have one.

Check your long-context requests. GPT-6's 2x / 1.5x surcharge above 272,000 input tokens can erase the price cut for document-heavy workloads. Claude's 1M window has no surcharge. Measure the input length of your real requests with the token calculator before choosing.

Use the cache. Both vendors now charge 5% to 10% of the input price for cached prefixes. Structure prompts so the static part (system prompt, tool definitions, reference documents) comes first and the variable part last, and the bulk of your input tokens become nearly free on repeat calls.

Do not assume the December prices. Google's Gemini 3.8 Flash promotional rate doubles on 1 January 2027, and OpenAI and Anthropic have both changed prices twice this quarter. Budget on today's list price, but re-check quarterly.

For a side-by-side of capabilities, context windows and prices across all of these models, read GPT-6 Sol vs Luna vs Astra vs Claude Opus 5.5 vs Gemini 3.8 Flash.

Sources

• TechCrunch, OpenAI launches GPT-6 Sol and Luna, 22 September 2026
• Anthropic, Introducing Claude Opus 5.5, 22 September 2026
• OpenAI, API pricing and model pages for gpt-6-sol, gpt-6-luna and gpt-6-astra
• Anthropic, Claude pricing documentation