GPT-6.1 Terra and Luna API Pricing: Cost Math for Content Teams

GPT-6.1 Terra, released on 26 October 2026, costs $1 per million input tokens and $5 per million output tokens, and GPT-6 Luna costs $0.10 and $0.50. For a content team drafting 200 articles a month, my worked example puts the model bill at about $3.70 on Terra and $0.37 on Luna, against $37 on GPT-6 Astra. Review time is the larger cost.

Key Takeaways

• OpenAI list prices per million tokens (input/output), as read in November 2026: Astra $10/$50, GPT-6.1 Sol $2/$10, GPT-6.1 Terra $1/$5, GPT-6 Luna $0.10/$0.50, per OpenAI's pricing page.
• Terra is priced 50% below GPT-5.6 Terra and GPT-6.1 Sol, and at one-tenth of Astra, per BleepingComputer.
• Output tokens drive the bill. In my 200-article example, 68% of Terra's cost is output, so shortening prompts saves little and capping output length saves more.
• Two rival prices are time-limited: Gemini 4 Argon doubles to $4/$20 on 1 December 2026, and Gemini 3.8 Flash doubles on 1 January 2027, per Google's pricing page.
• On OpenAI's own Agents' Last Exam, Claude Opus 5.5 scored higher than any GPT model listed, at a higher cost per task, per InfoWorld. Cheaper per token does not always mean better per task.
• Batch and Flex processing for OpenAI's GPT-6 models list at 50% of standard rates, which halves every figure below for work that can wait.

The price table

Prices as read from each vendor's page in November 2026, and they change. Prompts over 272K input tokens cost 2x input and 1.5x output on OpenAI's GPT-6 models, per the OpenAI page. GPT-5.6 Sol's promotional price runs to 30 November 2026 on the pricing page, though the API changelog says "at least" 21 November. I trust the pricing page, but verify before you commit.

What is new since my earlier cost posts: the Terra tier did not exist when I wrote about Sol for content teams or Claude Sonnet 5.5's cost math, and Terra is the tier that undercuts both.

Worked example 1: 200 drafts a month

These are my assumptions, not measurements from any team. Each job sends 6,000 input tokens (brief, style guide, source notes) and receives 2,500 output tokens (a draft of roughly 1,800 words). Volume is 200 jobs a month.

The arithmetic: Astra is 6,000 x $10 / 1M plus 2,500 x $50 / 1M, which is $0.06 plus $0.125.

What the example shows

• Even the top tier is cheap at this volume. $37 a month for 200 drafts is less than one hour of many editors' time. Compare it with your own rate, because I do not know yours.
• Output dominates. For Terra, output is $0.0125 of the $0.0185 per draft. If you cache the 4,000-token style guide at $0.05 per million, the per-draft cost drops from $0.0185 to $0.0147, about 20%. Useful, not dramatic.
• Sol, Sonnet 5.5 and Gemini 4 Argon (today) price identically at $2/$10. Choose among them on quality for your style, not price. My Sonnet cost post and Argon cost post have the detail. Argon's cost doubles in December.
• Luna and Haiku 5.5 price identically at $0.10/$0.50. The question is which one passes your checklist. See Haiku 5.5 versus Sonnet 5.5 for content ops.

Worked example 2: 2,000 short tasks

Meta descriptions or product blurbs: 800 input tokens, 150 output tokens, 2,000 jobs a month.

Under a dollar for Luna means the model is not the thing to optimize. The thing to optimize is how many of the 2,000 outputs need rewriting.

Cost per task is different from cost per token

Per-token price ignores how many tokens a model burns to reach an answer. OpenAI's own Terra benchmarks, as reported by InfoWorld, show the gap:

These are OpenAI-reported figures, and the same coverage notes Claude Sonnet 5.5 led OSWorld 2.1 at 83.9% against Terra's 74.2%. Two lessons. First, per-task cost does not track per-token price. Terra is 10x cheaper than Astra per token but about 7x cheaper per task, so Astra used fewer tokens to finish; against GPT-5.6 Terra the gap widens from roughly 2x per token to about 3.5x per task. Second, the highest score cost the most. An analyst quoted by InfoWorld said cheaper inference fixes only part of agent economics, since review and integration costs remain. I agree: price a task, not a token.

What this means for marketers and founders

• Start every structured chore on Luna or Haiku. Classification, tagging, title variants and schema drafts. Escalate only the outputs that fail your checklist. My cheap-model decision post explains when this beats frontier models.
• Use Sol or Sonnet-class models for anything with your byline. The price gap to Astra is five times, but the quality gap on ordinary marketing copy is something you have to measure, not assume.
• Test Terra on screen-driven work only if you already automate it. SERP audits and CMS form filling are the candidate tasks, per OpenAI's positioning of Terra for computer use.
• Put expiry dates in your budget sheet. Argon on 1 December, Gemini 3.8 Flash on 1 January 2027, GPT-5.6 Sol's promotion on 30 November.
• Run a 20-sample bake-off. Take 20 real briefs, run them through two tiers, and have one editor score blind. My protocol is here.

I would not rebuild a stable workflow because a cheaper tier appeared. Do it when the savings exceed the cost of re-testing, and for a content team spending under $50 a month on models, they usually do not.

What we do not know yet

• Terra's context window. None of the coverage I opened states it.
• Whether OpenAI's price cuts are permanent. The GPT-5.6 promotions had end dates, and nothing I read says the GPT-6 prices are promotional.
• How these models score on writing quality. The benchmarks above test agents and computer use, not brand-voice drafting.
• Real token counts for your workflow. Mine are assumptions, so replace them with your own logs.

FAQ

How much does GPT-6.1 Terra cost?

GPT-6.1 Terra costs $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens, per OpenAI's pricing page. That is about half of GPT-5.6 Terra's input price and one-tenth of GPT-6 Astra's. Batch and Flex processing list at half those rates.

How much does GPT-6 Luna cost?

GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens, with $0.01 for cached input. It is OpenAI's cheapest GPT-6 model and matches Claude Haiku 5.5's standard rates for prompts up to 100,000 tokens.

Is GPT-6 Astra worth ten times the price of Terra?

For routine content work, usually not. OpenAI's own figures show Terra scoring 60.0% against Astra's 59.3% on one agent benchmark at one-seventh the cost per task. For tasks needing the strongest coding or reasoning, OpenAI says Sol and Astra remain better.

Which is cheaper, GPT-6.1 Sol or Claude Sonnet 5.5?

They are identical on list price at $2 input and $10 output per million tokens. Cached input is also the same at $0.10. Differences come from how many tokens each model uses and the quality of its output for your task.

What will Gemini 4 Argon cost after 1 December?

Google's pricing page lists Gemini 4 Argon at $4 input and $20 output per million tokens from 1 December 2026, up from $2 and $10. Batch and Flex are half of standard rates in both periods.

Does prompt caching save much on content generation?

Less than people expect. In my 200-draft example, caching a 4,000-token style guide on Terra cuts cost per draft by about 20%, because output tokens make up most of the bill. Caching matters more for long, repeated prompts with short answers.

Should I use the Batch API for content?

If you can wait, yes. OpenAI's pricing page lists Batch at 50% of standard rates for GPT-6 models. It suits overnight jobs such as bulk metadata and not live editing sessions.

Want help with this?

I'm Younus Fardeen, an organic growth, SEO and AEO strategist with 4+ years of marketing experience across edtech and startup teams. If you want help costing and testing an AI content workflow for your team, see my work and reach me through the contact form at younusfardeen.com.