GPT-6 Sol vs GPT-6 Luna vs Claude Opus 5.5: The September 2026 Model Pricing Breakdown
The three models, same afternoon
September 22, 2026 gave developers three new models to price out on the same afternoon: OpenAI's GPT-6 Sol and GPT-6 Luna and Anthropic's Claude Opus 5.5, all of them cheaper in some meaningful sense than the tiers they relate to, and all of them aimed at different parts of the workload spectrum. That makes this the first September 2026 morning where the headline question for a lot of engineering teams was not "which lab just shipped the biggest number" but "GPT-6 Sol vs Claude Opus 5.5 — and where does the cheapest GPT-6 option, Luna, fit in?" Sol is OpenAI's mid-tier at $2 per million input tokens and $10 per million output tokens, built for multistep work like agentic coding and long-running tasks, with a 1,050,000-token context window and a maximum output of 128,000 tokens. Luna is the low tier at $0.10 in and $0.50 out, the cheapest GPT-6 model OpenAI has shipped, with the same 1,050,000-token context window and 128,000-token maximum output, and it is the one OpenAI is making available to Free and Go users in the ChatGPT desktop app as well as to paid accounts in ChatGPT Work, Codex, and the API. Opus 5.5 is Anthropic's first Claude 5.5 family model at $4 per million input and $20 per million output — 20% cheaper than Opus 5 on raw per-token pricing — which Anthropic says performs at the level of Fable 5.1 for most tasks while costing about 40% less to run on typical workloads, with a 1 million-token context window, multimodal input, and a same-day landing in GitHub Copilot for Pro+, Max, Business, and Enterprise subscribers.
On-paper pricing: Sol is the cheap flagship-adjacent option
The cheapest model on paper is Luna at $0.10 in and $0.50 out, and the cheapest flagship-adjacent model on paper is GPT-6 Sol at $2 in and $10 out, which is half the input price and a lower output price than GPT-6 Sol vs Claude Opus 5.5 would suggest on the Anthropic side at $4 in and $20 out. The on-paper per-token ranking is not the whole story, though, because Anthropic is leading with a typical-workload cost claim — about 40% less to run than Opus 5 — that includes token efficiency rather than just list price, and if Opus 5.5 spends fewer steps and tokens to finish the same work then the real bill for a given workload can come out well below what the per-token math alone suggests. That is the key distinction for anyone pricing a workload this week: Sol is the cheapest option for serious multistep work on raw per-token pricing among the three, Luna is the cheapest option full stop but is the focused high-volume tier rather than the model you pick for a complex multistep agent, and Opus 5.5 is the most expensive on paper but carries a workload-efficiency claim that could make it competitive on total cost depending on the job.
GPT-6 Sol: the OpenAI middle rung
GPT-6 Sol is the middle rung of OpenAI's new GPT-6 stack, sitting below the flagship GPT-6 Astra and above the low-tier Luna, and it is the model OpenAI is positioning for the serious middle of the workload spectrum — the multistep coding and agentic tasks that want GPT-6 reasoning without paying Astra prices for everything. The standard short-context pricing is $2 per million input tokens and $10 per million output tokens, with cached input reads at $0.20 per million and cache writes billed at 1.25x the uncached input rate, and prompts with more than 272,000 input tokens are priced at 2x input and cache rates and 1.5x output for the full request. Regional processing adds a 10% premium where available, and EU data residency is available only with Standard processing. The context window is 1,050,000 tokens with a maximum output of 128,000 tokens, and on Artificial Analysis's numbers Sol scores around 82.2 on the Decision benchmark and ranks fourth out of 196 models tracked, with its strongest eligible category in coding at number seven. It is available in ChatGPT Work and Codex for most paid accounts and in the ChatGPT API.
GPT-6 Luna: the access play
GPT-6 Luna is the low end of the stack and the one that changes the access picture the most, built for focused, high-volume work and pairing GPT-6 reasoning and tools with a price aimed at volume. The pricing is $0.10 per million input tokens and $0.50 per million output tokens, with the same 1,050,000-token context window and 128,000-token maximum output as Sol, and the distribution is what sets Luna apart — it is available in ChatGPT Work and Codex for most paid accounts, in the ChatGPT API, and also in the ChatGPT desktop app and for Free and Go users, which is the part that makes it a base-expanding move rather than simply a low API tier. The model is aimed at the high-volume focused task end of the spectrum rather than the complex multistep end, so the honest read is that Luna is the model you pick when the job is a lot of discrete, focused work rather than a long agentic chain.
Claude Opus 5.5: the Anthropic flagship, repriced
Claude Opus 5.5 is Anthropic's newest flagship and the first model in the Claude 5.5 family, positioned as the company's most advanced model to date while performing at the level of Fable 5.1 for most tasks and costing about 40% less to run than Opus 5 on typical workloads at default settings. The first-party API pricing is $4 per million input tokens and $20 per million output tokens — 20% cheaper than Opus 5 on input and output — with cached input reads at $0.20 per million, which is 60% cheaper than the comparable Opus 5 cache rate, and cache writes at $5 for a five-minute window. Batch processing is half price at $2 input and $10 output, and a fast mode is available on the first-party API at $8 input and $40 output. The context window is 1 million tokens with multimodal input, and Anthropic says the model leads on its own benchmarks in agentic coding, computer use, and knowledge work. Opus 5.5 landed in GitHub Copilot the same day it launched — September 22, 2026, with Anthropic publishing "Claude Opus 5.5 is available today" at 16:31 UTC and putting the announcement live at anthropic.com/claude-opus-5-5 the same minute — for Pro+, Max, Business, and Enterprise subscribers in a phased rollout with no universal availability date announced yet.
The real comparison: workload cost, not list price
The context windows are close enough that they are not the deciding factor for most workloads. Sol and Luna both offer 1,050,000 tokens, and Opus 5.5 offers 1 million tokens, so if you are routinely pushing past a million tokens the difference is real, but for most workloads it is not the thing you are optimizing for. The availability picture is where the three models diverge most: Sol and Luna are both in ChatGPT Work, Codex, and the API, with Luna also in the desktop app and for Free and Go users — a broad push down the access ladder — while Opus 5.5 is in the first-party API and just entered GitHub Copilot in a phased rollout, which means Anthropic's model is the one that lands inside a developer tool developers were already using the fastest.
If you are choosing between GPT-6 Sol vs Claude Opus 5.5 for a multistep coding or knowledge-work workload in September 2026, the honest short answer is that Sol is the cheaper option on raw per-token pricing and a strong option on coding according to third-party benchmarks, while Opus 5.5 is the model to use if you want Anthropic's top capability range and care about total-workload cost rather than list price — and it has the advantage of being inside GitHub Copilot the same day it launched. If your workload is high-volume and focused rather than complex and multistep, Luna is the model that changes the math most, and OpenAI is deliberately making it available to Free and Go users in a way that expands the base rather than simply lowering a price. The three models are not really competing in the same lane across the board — GPT-6 Sol and Claude Opus 5.5 overlap most directly on serious multistep work, while Luna occupies a cheaper, more focused lane that does not map neatly onto either flagship-adjacent model.
The broader read is that September 2026 is the month the frontier model catalog stopped being organized around "flagship versus everything else" and started being organized around "what does this specific workload cost at the tier that fits it." OpenAI built a three-tier GPT-6 stack and pushed the low tier into free accounts, and Anthropic made its flagship cheaper to run than the model it replaced and put it in Copilot the same day. The benchmark race did not stop — GPT-6 Astra is still the headline capability story from two weeks earlier, and the September 2026 benchmark round-up on Astra, Fable 5.1, and Gemini 3.8 is still the most complete picture of where the three frontier models stand — but the pricing-and-access race is the one that changes what most teams actually do, and that is the race both labs chose the same afternoon to run. For the wider industry context on why OpenAI and Anthropic both moved toward cheaper access on the same day, see our same-day launches breakdown, and for the Anthropic-specific read on what Opus 5.5 means for Fable 5.1 users, see our Claude Opus 5.5 flagship analysis.
— Sources: OpenAI's GPT-6 Sol and Luna announcement (openai.com, Sep 22 2026); Anthropic's Claude Opus 5.5 announcement (anthropic.com/claude-opus-5-5, Sep 22 2026); OpenAI API docs for GPT-6 Sol; OpenAI Developer Community announcement; GitHub Copilot changelog (github.blog, Sep 22 2026); OpenRouter; Artificial Analysis; BenchLM.ai; TechCrunch; The New Stack; Eigent AI; Coursiv; The Rundown; Kingy AI; CellCog; 9to5Mac; BNN Bloomberg; Yahoo Finance.
