OpenAI API pricing (2026): the GPT-6 per-token cost table
Short answer: Per million tokens, OpenAI's current GPT-6 family (released September 2026) costs $10 in / $50 out for GPT-6 Astra (top tier), $2 in / $10 out for GPT-6.1 Sol, and $0.10 / $0.50 for GPT-6 Luna — the cheapest tier. The previous flagship, GPT-5.6 Sol, runs $4 / $20 under promotional pricing through Nov 21, 2026 ($5 / $30 standard). Cached input saves ~90% and the Batch API takes 50% off. Verified against official OpenAI pricing.
OpenAI API model pricing (per 1M tokens)
| Model | Input | Output | Best for |
|---|---|---|---|
| GPT-6 Astra (top tier, Sept 2026) | $10 | $50 | Hardest end-to-end work; same price as Claude Fable 5.1 |
| GPT-6.1 Sol (Sept 29, 2026) | $2 | $10 | Best value for most production use |
| GPT-6 Luna (cheapest) | $0.10 | $0.50 | Routing, classification, high-volume simple tasks |
| GPT-5.6 Sol (previous flagship) | $4 (promo) | $20 (promo) | Previous-generation flagship |
| GPT-5.6 Terra | $2 | $12 | Previous-generation mid-tier |
| GPT-5.6 Luna | $0.20 | $1.20 | Previous-generation budget tier |
Prices are USD per million tokens (MTok) for standard synchronous calls. GPT-5.6 Sol's $4 / $20 rate is promotional pricing in effect through November 21, 2026; its standard rate is $5 input / $30 output. The GPT-5.6 tiers share a ~1.05M-token context window (1,050,000 tokens) with a 128K-token max output. Requests over 272,000 input tokens trigger a long-context premium on both generations (on GPT-6, input doubles and output rises 1.5x; e.g. GPT-6.1 Sol goes to $4 / $15), applied to the entire request. "ChatGPT API pricing" refers to these GPT-6 and GPT-5.6 models — the consumer ChatGPT product is a separate $20/mo subscription, not per-token billing.
What is the cheapest OpenAI API model?
GPT-6 Luna, at $0.10 / $0.50 per million tokens, is the cheapest current-generation OpenAI model — half the price of GPT-5.6 Luna ($0.20 / $1.20). The effective cost drops further with two stacking discounts:
- Cached input: repeated prompt prefixes are discounted ~90%, so long static system prompts cost a fraction of full input.
- Batch API: 50% off both input and output for asynchronous, non-time-sensitive jobs.
For tasks where even Luna is overkill, compare against Claude Haiku 4.5 and Gemini Flash in API alternatives.
Discounts: cached input and Batch API
| Feature | Discount | Applies to |
|---|---|---|
| Cached input | ~90% off input | Repeated prompt prefixes (e.g. Sol cached input ~$0.40–$0.50) |
| Batch API | 50% off in & out | Async jobs (24-hour window) |
| Flex / lower-priority | Reduced rate | Latency-tolerant workloads |
Worked example: cost of a typical GPT-5.6 Terra request
A request that sends 20,000 input tokens and generates 2,000 output tokens on Terra:
- Input: 20,000 × $2 / 1,000,000 = $0.040
- Output: 2,000 × $12 / 1,000,000 = $0.024
- Total: $0.064 per request
The same request on Luna costs (20,000 × $0.20 + 2,000 × $1.20) / 1,000,000 = $0.0064 — 10x cheaper, which is why Luna is the right pick for high-volume, low-complexity work.
How OpenAI API pricing compares to Claude
OpenAI still wins on raw price at the cheap end: GPT-6 Luna ($0.10/$0.50) undercuts Claude Haiku 4.5 ($1/$5) by a wide margin. In the middle the two are now priced identically — GPT-6.1 Sol and Claude Sonnet 5.5 both cost $2/$10. At the top, GPT-6 Astra matches Claude Fable 5.1 exactly at $10 input / $50 output, while Claude Opus 5.5 ($4/$20) sits in between. Full breakdown: ChatGPT vs Claude API cost and the Anthropic API pricing table.