Looking for GPT-6.1 Luna pricing? As of October 28, 2026, OpenAI’s model catalog lists GPT-6 Luna, identified as gpt-6-luna, rather than a GPT-6.1 Luna release. For developers budgeting a production workload, the available model starts at $0.10 per million input tokens and $0.50 per million output tokens through OpenAI. CometAPI publishes corresponding short-context rates of $0.08 and $0.40.
To decide whether to stay with GPT-6 Luna or move to GPT-6.1 Terra, look beyond the headline token rates. Prompt length, caching, reasoning, tool calls and retries all affect the cost of a finished task. This guide explains Luna pricing, compares Luna with Terra, and helps you choose where an upgrade is worth the additional spend.
Reviewed for clarity, source attribution and current API terminology.
Key Takeaways
Model identity: Use gpt-6-luna for GPT-6 Luna API requests; a GPT-6.1 Luna price is not listed in the official catalog.
Standard pricing: GPT-6 Luna costs $0.10 input / $0.50 output per million tokens through OpenAI, or $0.08 / $0.40 at CometAPI’s published short-context rates.
Context and billing: The model has a 1,050,000-token context window; prompts above 272,000 input tokens use higher rates for the full request.
Cost controls: Cache repeated input, limit unnecessary output, and choose reasoning effort that meets your acceptance criteria.
Model selection: Keep Luna for inexpensive routine work; evaluate Terra when visual interpretation, screen interaction, or writing quality affects successful completion.
Does GPT-6.1 Luna exist?
No separate GPT-6.1 Luna model is listed in the official API catalog checked on November 2, 2026. For Luna requests, use the documented model ID gpt-6-luna. A search for “GPT-6.1 Luna price” does not establish a released product or a separate rate card. “Fe-Pichu” is an OpenAI model available on Arena. It is highly likely to be Luna—given that Terra is already live and this model’s output quality falls below that of Terra—and the fact that it uses an OpenAI tokenizer further suggests it is an OpenAI model.
If you searched for that name because you want a newer alternative to Luna, consider GPT-6.1 Terra on CometAPI. OpenAI announced Terra on October 26, 2026, describing a focus on visual reasoning, computer use, and natural writing. For an existing Luna workflow, it is a candidate replacement when those capabilities address a recurring weakness. This is a workflow recommendation, not a claim that OpenAI renamed or discontinued Luna.
Start by identifying what needs to improve: fewer screen-reading errors, better image interpretation, or less editing after generation. Compare both models on those tasks before switching. If Luna already meets your quality and response-time targets, its lower token price remains a good reason to keep it.
What Is GPT-6 Luna?
OpenAI released GPT-6 Luna on September 22, 2026 for focused, high-volume work. It accepts text and images and produces text. Its practical appeal is economical automation: classification, extraction, and other bounded tasks whose results you can check consistently.
Luna offers configurable reasoning and a large context window, but a low token price alone does not make every task inexpensive. Failed attempts, long responses, and manual corrections all affect the final cost. Use it where the task is clear and the acceptance criteria are easy to apply.
For tasks that need more reliable interpretation of visuals, interaction with application screens, or more natural prose, GPT-6.1 Terra is a practical alternative to test. The choice is whether its results justify the extra spend in your workflow.
Where GPT-6 Luna Fits in the OpenAI Model Lineup
Choose a model around the work you need to complete and the budget you can sustain. In the current lineup, Astra covers demanding reasoning, Sol balances capability and cost, and Luna serves economical, high-volume automation. Terra adds a mid-priced option for visual tasks, screen interaction and writing.
Model
Position
When to consider it
Input / output per 1M
GPT-6 Astra
Reasoning flagship
Complex reasoning and coding when capability takes priority over token cost.
$10.00 / $50.00
GPT-6.1 Sol
Mid-priced general option
Work that needs a balance of model capability and spending.
$2.00 / $10.00
GPT-6.1 Terra
Mid-priced visual and writing option
Visual interpretation, application screens, and drafting or revising prose.
$1.00 / $5.00
GPT-6 Luna
Low-cost, high-volume option
Repeated, cost-sensitive tasks such as extraction and classification.
$0.10 / $0.50
Prices are USD per million tokens for OpenAI Standard short-context fresh input and output. Caching, longer prompts, other processing modes and additional services can change the total.
For an existing Luna application, test Terra on the tasks where visuals, screen actions or writing quality cause extra attempts. Keep Luna for routine requests that already meet your quality target, and compare the total cost of accepted results before moving more traffic.
GPT-6 Luna API Pricing
The official rates below separate fresh input, cached-input reads, cache writes, and output. All figures are USD per million tokens. Short context means a prompt with no more than 272,000 input tokens; longer prompts move the entire request to the long-context schedule.
OpenAI processing mode
Fresh input / 1M
Cached read / 1M
Output / 1M
Standard — short context
$0.100
$0.010
$0.500
Batch / Flex — short context
$0.050
$0.005
$0.250
Fast — short context
$0.200
$0.020
$1.000
Standard — long context
$0.200
$0.020
$0.750
The GPT-6 Luna billing specification applies 2× input and cache rates and 1.5× output rates above 272,000 input tokens. The 1,050,000-token context window is a capacity limit, not the boundary for the lower price tier. Standard 30-minute cache writes cost $0.125 per million short-context tokens; the caching example below shows how a write affects total savings.
Batch and Flex use half the Standard rates; Fast uses twice the applicable rates. Choose a processing mode based on timing and availability requirements, and check the selected provider’s support before applying a direct-API discount to gateway traffic.
CometAPI pricing note: The CometAPI rate card for GPT-6 Luna checked on October 28, 2026 lists short-context rates of $0.08 for fresh input, $0.008 for cached reads, $0.10 for 30-minute cache writes and $0.40 for output, per million tokens. Long-context fresh input and output are listed at $0.16 and $0.60. These rates are 20% below the corresponding OpenAI Standard rates shown here; processing-mode discounts and tool availability should be checked separately. For a short-context request with 5,000 fresh input tokens and 1,000 billed output tokens, the model-token cost is $0.00100 through OpenAI Standard or $0.00080 through CometAPI, excluding caching and separately billed services.
GPT-6 Luna reasoning effort and total spend
Reasoning effort changes how much model work a request may use; it is not an additional fixed percentage in these token tables. Compare effort settings on the same tasks, including generated reasoning, retries, and final-output usage. A shorter visible answer does not necessarily mean fewer billed output tokens.
To get started, follow the CometAPI quickstart to create an API key and configure the OpenAI-compatible base URL. Select gpt-6-luna and confirm model access and endpoint support for your account before integrating it into a production workflow.
How Much Performance Are You Buying With GPT-6 Luna?
OpenAI’s published benchmark results provide evidence of Luna’s cost-performance trade-off. These are provider-reported evaluations, not independent CometAPI tests. Compare the benchmark’s task set and agent configuration with your own workload before using its cost savings in a forecast.
GPT-6 Luna benchmark results and test conditions
DeepSWE v1.1: Luna scored 66.6% at max reasoning effort, compared with 68.8% for GPT-6 Sol at max. Luna’s reported task costs were 93% below Claude Opus 5 and 96% below Claude Fable 5, with both competing models evaluated at medium effort in those comparisons.
AutomationBench 1.0.6: At high effort, Luna improved on GPT-5.6 Luna by 5.4 percentage points at 58% lower cost per task. The evaluation covers end-to-end business workflows using 47 tools. These savings describe that evaluation, rather than a guaranteed reduction for every application.
OSWorld 2.0: Luna at max exceeded GPT-5.6 Sol at medium at about one-tenth of the task cost. The reported metric is partial reward on the offline set from release v2026.08.08. The announcement does not provide every agent and runtime detail needed to reproduce these results.
The chart plots task cost against benchmark score. Its Claude Fable 5.1 series omits Opus 5 fallback costs on about 40% of tasks, so the plotted cost understates that configuration’s total. OpenAI’s announcement explains the evaluation; the chart is reproduced in Soft4fun’s launch coverage.
For deployment, measure cost per accepted task: a lower token price can be offset by repeated attempts, tool execution, or human correction.
How Does Prompt Caching Change GPT-6 Luna Input Costs?
The OpenAI prompt-caching guide explains how reuse of a stable prefix can reduce input processing. A cached-read discount applies to reused input, not to the entire request. Count writes, reads, and ordinary input separately.
Operation
OpenAI Standard short-context price / 1M
Fresh input
$0.100
30-minute cache write
$0.125
Successful cached read
$0.010
Suppose ten requests reuse the same 100,000-token prefix. Assume one 30-minute write and nine successful reads before expiry, with no additional writes. The prefix remains below the long-context threshold and each request leaves room for other input and output.
Repeated-prefix component
Without caching
One write + nine reads
Ordinary prefix input
10 × 100,000 × $0.10 / 1M = $0.1000
$0
Initial write
$0
100,000 × $0.125 / 1M = $0.0125
Nine reads
$0
9 × 100,000 × $0.01 / 1M = $0.0090
Repeated-prefix total
$0.1000
$0.0215
The prefix saving is 78.5%. If each request also produces 2,000 billed output tokens, output adds $0.0100 across the ten requests. Model-token totals become $0.1100 without caching and $0.0315 with caching, a saving of about 71.4% before other input and services.
A repeated prompt is not proof of a cache hit. Check measured cached usage, write retention, and prefix stability. The example uses OpenAI direct rates; estimate a CometAPI route with that route’s cache schedule and usage reporting.
GPT-6 Luna vs GPT-6.1 Terra: Which Should You Choose?
For teams already using GPT-6 Luna, GPT-6.1 Terra is a candidate replacement when the task moves beyond straightforward volume processing. Choose based on what improves the finished result, rather than the version number.
Price: GPT-6 Luna retains the lower token cost
According to the GPT-6 Luna model page and GPT-6.1 Terra model page, OpenAI Standard short-context rates per million tokens are $0.10 input and $0.50 output for GPT-6 Luna, versus $1.00 input and $5.00 output for GPT-6.1 Terra. Fresh-input and output rates are therefore ten times higher for GPT-6.1 Terra.
A request with 5,000 fresh input tokens and 1,000 billed output tokens costs $0.001 on GPT-6 Luna or $0.010 on GPT-6.1 Terra at those rates. This example excludes caching, tools, retries, and longer-context adjustments. If either model needs extra turns, compare the complete run.
Task fit: keep GPT-6 Luna for routine work; test GPT-6.1 Terra for visual and writing tasks
GPT-6 Luna is a sensible starting point for repeated extraction, routing, and classification with clear validation rules. Test GPT-6.1 Terra when interpreting screenshots or visual documents, acting on application screens, or revising prose is central to the job. OpenAI's model-selection guidance highlights GPT-6.1 Terra for concise writing, defined computer-use tasks, and scoped code fixes.
Replacement decision: measure successful results
Use a representative set of real requests and the same acceptance criteria. Record success rate, end-to-end response time, retries, and any human editing. Divide the full run cost by accepted results. Move the tasks where GPT-6.1 Terra delivers a worthwhile improvement, and keep GPT-6 Luna on the routine work it already handles well.
You can evaluate both through GPT-6 Luna's CometAPI page and GPT-6.1 Terra's CometAPI page. The latter currently lists $0.80 input and $4.00 output per million tokens. Use the chosen provider's current prices and supported endpoints in your comparison; these are separate from OpenAI direct rates.
How Can You Reduce API Costs Further?
Use prompt caching for repeated content
Reusable system instructions, reference documents, and schemas can benefit from caching. The OpenAI caching guide explains how prompt reuse can reduce fresh-input work. For Luna, eligible cached-input reads cost 90% less than uncached reads, but cache writes have their own rates.
Match reasoning effort to the job
Start with the least expensive effort setting that meets a quality threshold. Simple routing and extraction frequently need less reasoning than code debugging or multi-step analysis. Higher effort may increase reasoning-token consumption and the total bill.
Keep responses appropriately short
Avoid asking for verbose explanations if the application only requires a category, three fields, or structured JSON. Output tokens are substantially more expensive than input tokens at Standard rates.
Route difficult requests selectively
A practical architecture starts with Luna and escalates failed validation, ambiguous requests, or complex reasoning to a stronger model. Evaluate the additional model cost against accepted-task improvements rather than assuming all difficult cases require the premium tier.
GPT-6 Luna Alternatives: Fast Models to Test for Value
A useful GPT-6 Luna alternative should meet your response-time and quality targets at an acceptable total cost. The models below are candidates to evaluate, rather than a ranking of measured speed or a promise that each is cheaper than GPT-6 Luna.
GPT-6.1 Terra
Start with GPT-6.1 Terra on CometAPI when your main reason to move from GPT-6 Luna is visual work, screen interaction, or writing quality. It belongs on a shortlist for improving those tasks, although its token rates exceed GPT-6 Luna's. Keep the same test cases so that any improvement is easy to assess.
DeepSeek V4.1 Flash
DeepSeek V4.1 Flash on CometAPI provides another efficiency-focused option. The listed base rates are $0.12 input and $0.48 output per million tokens, with request-based pricing adjustments also shown on the page. Compare the rate that actually applies to your requests, then test structured output and completion time on your own workload.
Claude Haiku 5.5
Claude Haiku 5.5 on CometAPI is another model to include in a short pilot for everyday text and image-input tasks. Test its response style, adherence to your output format, and end-to-end latency before committing. Check its live rate card; inclusion here does not mean it undercuts GPT-6 Luna's price.
Conclusion
If you arrived looking for GPT-6.1 Luna, begin with the available model names: GPT-6 Luna for inexpensive routine automation, and GPT-6.1 Terra as an alternative to evaluate when visual work, computer interaction, or writing quality needs to improve.
Keep GPT-6 Luna where it meets your requirements. Introduce GPT-6.1 Terra where a better result reduces failed attempts or editing enough to justify the premium. Use the actual provider rates, full request history, and accepted results to make the decision. CometAPI's model pages provide a starting point for checking access and pricing before you run the comparison.
FAQ
Can I predict GPT-6 Luna cost from word count?
Use word count only for a rough budget. Token usage varies with language, formatting, code, and images, while generated reasoning can add billed output beyond the visible response. For a production estimate, sample representative requests and record the API’s measured usage; use the high end of that distribution when setting a spending limit.
Why can two similar GPT-6 Luna requests have different bills?
Check prompt length, cached-input eligibility, reasoning consumption, processing mode, and tool calls. A prompt that crosses 272,000 input tokens also changes the applicable rates for the full request. Compare those billing dimensions before treating a difference as an error.
How should I budget for a GPT-6 Luna fallback route?
Track how often requests escalate and whether the fallback processes the full original input. If 10% of tasks escalate, budget the first Luna attempt for every task plus the fallback requests for that 10%; do not replace the Luna cost with the fallback cost. Add retries and tools separately, then divide the total by accepted results.
Does a GPT-6 Luna cost estimate include ChatGPT access?
No. A token-based API estimate covers the assumptions in that calculation. It does not purchase a ChatGPT subscription or establish a plan’s model access, usage limits, or workspace eligibility. Keep the application’s API budget separate from employee subscription spending
