AI Tools
GPT-5.6 Luna pricing: what does the cheapest GPT-5.6 model cost?
GPT-5.6 Luna pricing: OpenAI charges $0.20 per 1M input tokens and $1.20 per 1M output tokens on its standard tier, verified on the official pricing page on August 28, 2026. That makes Luna the cheapest GPT-5.6 model, at 1/20th of GPT-5.6 Sol's input rate.
What matters
- GPT-5.6 Luna costs $0.20 per 1M input tokens and $1.20 per 1M output on the standard tier, with a 1.05M-token context window, verified August 28, 2026.
- OpenAI cut Luna's price 80% on July 30, 2026, per the API changelog; Terra dropped 20% the same day.
- Luna shares Sol's context window, 128K max output, and tool set (functions, web search, file search, computer use) at 1/20th of Sol's input price.
- Batch and Flex tiers halve Luna to $0.10 in and $0.60 out; Fast mode doubles standard rates for up to 2.5x faster responses.
- Segment co-founder Calvin French-Owen reports his personalized-news agent cost about $1 per run with Sonnet-class models and about $0.10 per run with Luna.
- Since August 6, 2026, Luna is also the default ChatGPT model for Free users, with unlimited text chats and a Think button for harder questions.
What is GPT-5.6 Luna?
GPT-5.6 Luna is OpenAI's cost-optimized model in the GPT-5.6 family, released on July 9, 2026 alongside GPT-5.6 Sol and GPT-5.6 Terra. OpenAI's model catalog describes it as "optimized for cost-sensitive workloads," with a 1.05M-token context window, 128K max output, and a knowledge cutoff of February 16, 2026.
Luna supports the same tool set as its siblings (functions, web search, file search, and computer use) and the full reasoning range from none to max, per the model documentation. It is available through the Responses API and OpenAI's SDKs.
Luna also sits inside ChatGPT. Since August 6, 2026 it is the default model for Free users, with unlimited text chats and a Think button that unlocks higher reasoning, per OpenAI's announcement. Our GPT-5.6 Sol vs Luna comparison covers the consumer plan split in detail.
The model drew wider developer attention this week when Segment co-founder Calvin French-Owen published "Small Models Have Arrived" on August 26, 2026. He reports running Luna across his codebase, email, and knowledge base at roughly 100 tokens per second, calling it "shockingly capable, fast, and smart," and argues that cheap small models finally make consumer AI products economically viable.
GPT-5.6 Luna pricing: the full rate card
OpenAI's pricing page, checked August 28, 2026, lists GPT-5.6 Luna at $0.20 per 1M input tokens, $0.02 per 1M cached input tokens, $0.25 per 1M cache writes, and $1.20 per 1M output tokens on the standard tier with short context. Prompts above 272K tokens, the long-context boundary named in OpenAI's August 5 changelog, double those rates to $0.40 in and $1.80 out.
The Batch and Flex tiers both halve the standard price to $0.10 in and $0.60 out per 1M tokens. Fast mode, which replaced the former Priority Processing tier on July 30, costs double the standard rate ($0.40 in, $2.40 out) and delivers up to 2.5x faster responses than the standard tier, per the changelog.
Two date-stamped facts matter for anyone budgeting: OpenAI cut Luna's price 80% on July 30, 2026 (Terra fell 20% the same day), and data-residency endpoints add a 10% uplift for models released on or after March 5, 2026.
GPT-5.6 Luna vs GPT-5.6 Sol and Terra
Inside OpenAI's own lineup, Luna is the budget end of a three-step ladder. GPT-5.6 Sol, positioned for "complex professional work," charges $4.00 in and $20.00 out per 1M tokens. GPT-5.6 Terra, which "balances intelligence and cost," charges $2.00 in and $12.00 out. Luna charges $0.20 in and $1.20 out.
That is a 20x input gap and a roughly 17x output gap between Luna and Sol, and 10x on both sides versus Terra. All three share the same 1.05M context window, the same 128K max output, and the same tool set, so the price gap is about capability headroom rather than features.
Pick Sol when the task needs frontier reasoning or complex coding. Pick Terra for a middle ground. Pick Luna for high-volume, cost-sensitive workloads where a cheaper token is worth a lower capability ceiling.
What does GPT-5.6 Luna actually cost to run?
Token price stays abstract until you multiply it by a real workload. French-Owen reports two concrete figures from his own usage. His pet project, a daily personalized news site that researches him across HN, Reddit, and Twitter, cost about $1 per run with the previous generation of Sonnet-class models. With Luna, the average run is about $0.10.
He also reports that complex research threads searching thousands of emails land in the tens of cents of API cost, and that it is difficult to run up a large bill even with long agent sessions.
Those are one practitioner's reports, not benchmarks, but they illustrate the point: at $0.20 per 1M input tokens, the per-request cost that blocked consumer AI economics starts to disappear. That is the essay's central claim and the reason the model is worth evaluating now.
The small-model landscape: GLM-5.3-Flash and DeepSeek V4 Flash
Luna is not the only cheap model at the frontier. French-Owen calls GLM 5.3 "a new option at the Pareto frontier" for capability per dollar. Zhipu's official pricing page, checked August 28, 2026, lists GLM-5.3 at ¥8 per 1M input tokens and ¥28 per 1M output, and GLM-5.3-Flash at ¥0.4 in and ¥1.4 out during a two-week half-price promotion (regularly ¥0.8 and ¥2.8). Both carry a 1M-token context window, and GLM-5.3-Flash also accepts image and video input.
DeepSeek's V4 Flash, verified on its API pricing page the same day, charges $0.22 per 1M input tokens (cache miss) and $0.66 per 1M output tokens off-peak, with cached input at $0.007. Its input price lands next to Luna's $0.20, while its output is about half of Luna's $1.20.
Each vendor prices differently: Zhipu in CNY, DeepSeek with peak and off-peak tiers, OpenAI with context-length and service tiers. The DeepSeek API pricing and GLM-5.3 pricing pages keep the full rate tables current.
Who should use GPT-5.6 Luna
- Agent builders: long prompts, tool calls, and loops reuse context; Luna's $0.02 cached input and $1.20 output keep agent sessions cheap.
- Consumer and SaaS apps: per-request AI features become viable at $0.20 per 1M input, the price point French-Owen argues unlocks consumer AI.
- Batch and background jobs: the $0.10 in / $0.60 out Batch tier is built for exactly this.
- Prototyping: the full reasoning range (none to max) and the same tools as Sol let you build against the family and scale up later.
Skip Luna for the hardest coding and novel reasoning. French-Owen says he still reaches for the most expensive, most capable models (Fable 5 and GPT-5.6 Sol) for coding work, and OpenAI positions Sol as the frontier choice for complex professional work.
The honest catch
Everything in this article traces to OpenAI's pricing page, model catalog, and changelog, checked August 28, 2026, plus French-Owen's essay, Zhipu's pricing page, DeepSeek's API pricing page, and OpenAI's August 6 ChatGPT announcement. We have not run Luna hands-on, and OpenAI publishes no benchmark suite for it, so treat capability claims as vendor positioning plus one practitioner's report.
Luna is a small model by design, and pricing can move. OpenAI already changed these rates once this year (the July 30 cut), GPT-5.6 Sol's promotional pricing runs only through November 21, 2026, and Zhipu's Flash discount is limited to two weeks. Re-verify the rate card before committing a workload to any of them.
At a glance
| GPT-5.6 Luna | GPT-5.6 Terra | GPT-5.6 Sol | |
|---|---|---|---|
| Positioning | Cost-sensitive workloads | Balances intelligence and cost | Frontier complex work |
| Standard input (short context) | $0.20 | $2.00 | $4.00 |
| Standard output (short context) | $1.20 | $12.00 | $20.00 |
| Cached input | $0.02 | $0.20 | $0.40 |
| Long-context input (over 272K) | $0.40 | $4.00 | $8.00 |
| Long-context output | $1.80 | $18.00 | $30.00 |
| Batch input / output | $0.10 / $0.60 | $1.00 / $6.00 | $2.00 / $10.00 |
| Fast mode input / output | $0.40 / $2.40 | $4.00 / $24.00 | $8.00 / $40.00 |
| Context window | 1.05M | 1.05M | 1.05M |
| Max output | 128K | 128K | 128K |
| Tools | Functions, web search, file search, computer use | Functions, web search, file search, computer use | Functions, web search, file search, computer use |
FAQ
Is GPT-5.6 Luna free?
In ChatGPT, yes: since August 6, 2026, GPT-5.6 Luna is the default model for Free users, with unlimited text chats. As an API, no: OpenAI lists no free tier, and standard rates are $0.20 per 1M input tokens and $1.20 per 1M output. The Batch and Flex tiers halve that to $0.10 and $0.60.
GPT-5.6 Luna vs GPT-5.6 Sol: which should I use?
Use Luna for cost-sensitive, high-volume workloads at $0.20 in and $1.20 out per 1M tokens. Use Sol for complex professional work at $4.00 in and $20.00 out. Both share a 1.05M context window, 128K max output, and the same tool set.
How fast is GPT-5.6 Luna?
OpenAI does not publish a speed spec. Calvin French-Owen reports roughly 100 tokens per second in his August 26 essay, and Fast mode (double the standard price) delivers up to 2.5x faster responses than the standard tier, per OpenAI's August 5 changelog.
Related reading
GPT-5.6 Sol vs Luna, DeepSeek API pricing, GLM-5.3 pricing, radar index