OpenAI API Pricing (November 2026): GPT-6 Astra, GPT-6.1 Sol, Terra and Luna Costs, Batch, Fast Mode and Tools

OpenAI’s API price list looks very different from a year ago. The GPT-6 family now has four tiers named after celestial bodies (Astra, Sol, Terra and Luna), “Priority” processing was renamed Fast mode on July 30, 2026, a new Ultrafast tier sits above it, and OpenAI is winding down its fine-tuning platform. If you are trying to budget an app, the naming alone is confusing.

I pulled every number below from OpenAI’s official API pricing page on November 5, 2026, and cross-checked the flagship prices against independent trackers. Prices are in US dollars per million tokens unless I say otherwise.

GPT-6 Model Prices

These are the four flagship text models on OpenAI’s pricing page, at the Standard processing tier.

Two details are worth noticing. First, cached input is extremely cheap on Sol, Terra and Luna, at 5% to 10% of the normal input price, so apps with a long shared system prompt benefit a lot. Second, GPT-6 Astra launched on September 3, 2026 with a context window of about 1.05 million tokens and 128K maximum output, according to independent model trackers.

OpenAI also lists separate configurations for special fields. GPT-6 Astra for Law (gpt-6-astra-law) costs the same $10 / $1 / $50 with promotional pricing through at least December 31, 2026. The cyber models gpt-5.6-sol ($4 / $0.40 / $20) and gpt-5.6-cyber ($12.50 / $1.25 / $75) and the life sciences models such as gpt-6-rosalind ($15 / $1.50 / $75) are for approved trusted-access programs only. OpenAI says GPT-5.6 Sol’s promotional pricing runs until November 30, 2026.

Batch, Flex, Fast Mode and Ultrafast

OpenAI now sells the same models at four speeds. You choose a tier per request.

• Batch API: half price in exchange for results within 24 hours. Perfect for bulk jobs.
• Flex processing: charged at Batch rates but runs through the normal API, with slower responses and occasional 429 errors when capacity is short. Good for background agents that can tolerate a wait.
• Fast mode: the old Priority tier, renamed on July 30, 2026. Requests still tagged as priority are routed to Fast automatically.
• Ultrafast: the quickest tier, available for Astra and Sol only, at six times the standard price.

Cached input discounts apply on top of each tier, so cached Batch input on Luna works out at just $0.005 per million tokens.

Codex, chat-latest and Specialised Models

Note that chat-latest is more expensive than GPT-6.1 Sol and its behaviour changes as ChatGPT updates, so use a pinned model for production apps. Embedding prices are from an independent tracker updated November 1, 2026.

Fine-tuning: OpenAI says it is winding down the fine-tuning platform and it is no longer available to new users. Existing fine-tuned gpt-4.1 models still run at $3 / $0.75 / $12, and gpt-4.1-mini at $0.80 / $0.20 / $3.20.

Image Generation Prices

The current image models are gpt-image-2.6-sunburst, gpt-image-2.6-flare and their 2.5 versions. All share the same token pricing:

The cost of each picture depends on size and quality, because larger images produce more output tokens. OpenAI points to the calculator in its image generation guide for exact per-image estimates.

Realtime Voice, GPT-Live and Transcription

For a voice assistant, GPT-Live at $0.05 a minute (model and tool usage billed separately) is the simplest to budget. A 10-minute call costs $0.50 for the session.

Built-in Tool Prices

Data residency endpoints add a 10% uplift for models released on or after March 5, 2026, and FedRAMP endpoints carry the same 10% uplift.

Usage Tiers and Rate Limits (Build, Launch, Grow)

On October 6, 2026 OpenAI replaced its old five paid usage tiers with three: Build, Launch and Grow. Your tier decides how much you can spend per month and how fast you can send requests. You move up automatically once your total credit purchases reach the threshold.

Two things people get wrong: the Free tier is a usage cap, not $100 of free credit, and the monthly caps are ceilings, not budgets. Ultrafast has its own separate limits. Check Settings > Limits in your OpenAI dashboard for the exact numbers on your account, because OpenAI can adjust them per organisation.

How to Get an API Key and Set Up Billing

1. Go to platform.openai.com and sign in (your ChatGPT login works, but billing is separate).
2. Open Settings > Billing, add a card and buy prepaid credits. $5 is enough to unlock the Build tier.
3. Turn on auto recharge only if you also set a limit, so a runaway script cannot drain your card.
4. Open Settings > Limits and set a monthly budget and an email alert threshold (for example, alert at $20, hard stop at $50 while testing).
5. Create a key under API keys. Give it a name, limit it to one project, and copy it once. You cannot view it again.
6. Store the key in an environment variable called OPENAI_API_KEY, never in your code or a public repo.

Then make a test call with the cheapest model so a mistake costs fractions of a cent:

curl https://api.openai.com/v1/responses \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-6-luna", "input": "Say hello in five words."}'

The response includes a usage object with input and output token counts, which is exactly what you are billed for.

How to Calculate Your Cost Before You Build

The formula is simple: (input tokens / 1,000,000 x input price) + (output tokens / 1,000,000 x output price). As a rule of thumb, 1,000 tokens is about 750 English words, so a one-page prompt is around 600 to 800 tokens.

Here is a tiny Python helper you can drop into a notebook to compare models for your own traffic:

PRICES = {  # USD per 1M tokens: (input, cached input, output)
    "gpt-6-astra":   (10.00, 1.00, 50.00),
    "gpt-6.1-sol":   (2.00, 0.10, 10.00),
    "gpt-6.1-terra": (1.00, 0.05, 5.00),
    "gpt-6-luna":    (0.10, 0.01, 0.50),
}

def monthly_cost(model, requests, in_tok, out_tok, cached_share=0.0, batch=False):
    p_in, p_cache, p_out = PRICES[model]
    total_in = requests * in_tok
    cached = total_in * cached_share
    cost = ((total_in - cached) * p_in + cached * p_cache + requests * out_tok * p_out) / 1_000_000
    return cost * 0.5 if batch else cost

for m in PRICES:
    print(m, round(monthly_cost(m, 30_000, 3_000, 400, cached_share=0.8), 2))

With the support-bot numbers used below and 80% caching, this prints about $852 for Astra, $163 for Sol, $82 for Terra and $8 for Luna per month. Reasoning models also produce hidden reasoning tokens that are billed as output, so add a margin of 20% to 50% on output if you use high reasoning effort.

OpenAI vs Other Providers at a Glance

Full details, including caching and batch for every provider, are in my AI API pricing comparison, plus the Gemini, Grok and DeepSeek pricing guides.

Billing and Rate Limit Errors (Fixes)

• 429 insufficient_quota: you have no credit left or hit your monthly budget. Add credits or raise the limit in Settings > Limits. Retrying will not help.
• 429 rate_limit_exceeded: too many requests or tokens per minute for your tier. Retry with exponential backoff, spread requests out, or move up a tier by buying more credits.
• 429 on Flex: Flex had no spare capacity. These failed requests are not billed. Retry later or use Standard.
• 401 invalid_api_key: the key was revoked, copied wrongly or belongs to another project. Create a new one.
• Bill higher than expected: check the Usage page by model. The usual causes are reasoning tokens, web search calls at $10 per 1,000, a missing cache (dynamic text placed before static text) or Fast mode left on.

What Changed in 2026

• New GPT-6 naming. Astra (top), Sol (mid), Terra (efficient) and Luna (smallest) replace the old mini and nano labels.
• Priority became Fast mode on July 30, 2026, and Ultrafast was added for Astra and Sol.
• Fine-tuning is winding down and closed to new users.
• The Assistants API shut down on August 26, 2026 according to independent trackers. The Responses API is the replacement.
• Older transcription models gpt-4o-transcribe and gpt-4o-mini-transcribe are deprecated, with shutdown reported for February 26, 2027.

Cost Examples

Same workload as in my Claude guide, so you can compare: a support bot with 30,000 conversations a month, 3,000 input and 400 output tokens each (90M input, 12M output).

If 80% of each prompt is a shared, cached system prompt, GPT-6.1 Sol drops to about $163 a month (72M cached tokens at $0.10, 18M fresh at $2, plus $120 output). These are estimates from list prices, not measured bills.

How to Lower Your OpenAI Bill

1. Start with Terra or Luna and only move up to Sol or Astra when the quality is not enough.
2. Put static content first in every prompt so it is served from the cache at 90% to 95% off.
3. Use Batch or Flex for anything that does not need an instant answer.
4. Avoid Fast and Ultrafast unless latency really earns you money.
5. Watch tool calls. Web search at $10 per 1,000 calls adds up fast in agent loops.

Frequently Asked Questions

What is the cheapest OpenAI model?

GPT-6 Luna at $0.10 input and $0.50 output per million tokens, or half that with Batch.

Is the OpenAI API included in ChatGPT Plus?

No. ChatGPT subscriptions and the API are billed separately, although ChatGPT Pro tiers come with some monthly API credits. See ChatGPT Free vs Go vs Plus vs Pro.

How much is GPT-6 Astra per 1,000 tokens?

$0.01 per 1,000 input tokens and $0.05 per 1,000 output tokens at the Standard tier.

Is OpenAI cheaper than Claude?

At the top end, GPT-6 Astra ($10 / $50) costs more than Claude Opus 5.5 ($4 / $20). GPT-6.1 Sol and Claude Sonnet 5.5 have the same list price ($2 / $10). Compare all providers in my AI API pricing comparison.

What is Flex processing?

A beta option billed at Batch rates that runs through the normal API, with slower answers and occasional 429 errors that are not billed.

Is there a free OpenAI API tier?

Yes, in eligible countries there is a Free tier with a $100 monthly usage cap and lower rate limits. It is a cap, not free credit. Buying $5 of credits moves you to Build.

What is the minimum I need to pay to start?

$5 of prepaid credits unlocks the Build tier, with a $500 monthly ceiling and 5,000 requests per minute.

How do I stop my API bill from running away?

Set a monthly budget and alert in Settings > Limits, use a separate key per project, avoid auto recharge without a limit, and log the usage field of every response.

How many words is a million tokens?

About 750,000 English words, roughly ten average novels. Code and non-English text use more tokens per word.

Do reasoning tokens cost extra?

They are billed at the output price even though you do not see them. Lower the reasoning effort for simple tasks to save money.

Final Thoughts

OpenAI’s 2026 price list rewards people who think about speed and caching. The same GPT-6.1 Sol request can cost $1, $2, $4 or $12 per million input tokens depending on the tier you pick, and cached input is almost free. For most apps I would start on GPT-6.1 Terra with caching, use Batch for background jobs, and keep Astra for the few tasks that truly need it.

To compare with the alternatives, read my Claude API pricing, Gemini API pricing and Grok pricing guides.

Sources

• OpenAI: API pricing (official)
• AI Business Weekly: OpenAI API pricing (updated Nov 1, 2026)
• Second Talent: GPT-6 vs Claude vs Gemini specs
• AI Weekly: GPT-6.1 Terra launch
• MIXED News: OpenAI cuts its API usage tiers from five to three (Oct 2026)