OpenAI API Pricing (October 2026): Which GPT-6 Tier to Use, and Every Discount

Quick answer: OpenAI bills the API per million tokens. Input, cached input and output each have their own rate, and every model has its own price. GPT-6 Astra is the premium tier, GPT-6.1 Sol is the near-Astra mid tier, and GPT-6 Luna is the budget tier. Batch and Flex cost half the standard rate, and Fast and Ultrafast cost more. Today’s numbers are on the official OpenAI pricing page.

Which OpenAI model tier should you use?

Tier	Best for	Key differences	Official pricing
gpt-6-astra	Hardest reasoning, agents, high-stakes output	Premium tier, the most expensive per token	OpenAI pricing
gpt-6.1-sol	Most production apps	Near-Astra quality at a fraction of Astra’s price; bigger cache discount than 6 Sol	OpenAI pricing
gpt-6-luna	Classification, extraction, high volume	Budget tier, the cheapest GPT-6 model	OpenAI pricing
gpt-5.6 family	Existing integrations	Older generation, on promotional pricing for now	OpenAI pricing
gpt-4o / gpt-4o-mini	Legacy apps	Still listed; usually worse value than GPT-6 Luna or Sol	OpenAI pricing

Requests above 272K tokens move to a higher long-context rate (up to about double the short-context rate).

Why did GPT-6 get cheaper in September?

When GPT-6 Sol and Luna launched, OpenAI said it had “made caching and inference more efficient” (@OpenAI, Sep 22). A week later it called GPT-6.1 Sol “near-Astra intelligence for a fifth of the price” (@OpenAI, Sep 29). Artificial Analysis noted that 6.1 Sol kept 6 Sol’s price but got a deeper cache-read discount (@ArtificialAnlys).

How do Batch, Flex, Fast and Ultrafast change the price?

Mode	Best for	Effect vs standard rate	Official pricing
Standard	Interactive apps	Base rate	OpenAI pricing
Batch API	Overnight and bulk jobs	Half price, asynchronous	OpenAI pricing
Flex processing	Jobs that can wait in a queue	Half price, slower	OpenAI pricing
Fast	Latency-sensitive features	2x the standard rate	OpenAI pricing
Ultrafast	When speed earns money	Premium; several times the standard rate	OpenAI pricing
Data residency	Regional compliance	+10%	OpenAI pricing

OpenAI says Ultrafast delivers “up to 6x” faster generation in the API (@OpenAI), and it reached GPT-6.1 Sol in the API on October 8 (@OpenAIDevs).

How are images and built-in tools billed?

• Image models (gpt-image-2, gpt-image-2.5, gpt-image-1-mini) are billed in image tokens, not per picture, so cost depends on size and quality. gpt-image-1-mini is the budget option.
• Web search is billed per 1,000 tool calls, plus the tokens used.

Rates for both are on the official pricing page.

How do OpenAI usage tiers and rate limits work now?

On October 6 OpenAI cut its paid tiers from five to three and made the top tier easier to reach (@OpenAIDevs). Per the rate-limits guide:

Tier	Who it fits	Key differences	Official page
Free	Testing from a supported country	Low monthly usage cap	Rate limits
Build	First paid projects	Unlocked by a first credit purchase; higher RPM/TPM	Rate limits
Launch	Apps with real traffic	Higher payment threshold and monthly cap	Rate limits
Grow	Scaling products	Highest tier and monthly cap	Rate limits

If you hit a hard spend limit you get a 429, and retrying won’t fix it. See fixing 401, 429 and quota errors.

How do I estimate my monthly OpenAI bill?

cost = input_tokens / 1e6 × input_rate
     + cached_tokens / 1e6 × cached_rate
     + output_tokens / 1e6 × output_rate

Plug in your expected token counts and the current rates from the pricing page. Output tokens cost several times more than input on every GPT-6 tier, so shorter answers save the most.

How can I cut OpenAI API costs?

• Keep the prompt prefix stable. Cached input is a small fraction of the normal input rate. OpenAI’s Codex lead has said a worse cache hit rate makes usage drain faster (@thsottiaux).
• Route by difficulty. Send simple tasks to Luna, and keep Sol or Astra for hard reasoning.
• Batch anything that isn’t user-facing for half price.
• Stay under 272K tokens per request to avoid the long-context rate.

Comparing providers? See Claude API pricing and DeepSeek API pricing, or go back to the AI API hub.