The same model at five prices.
A model's price used to be one number. It is now a starting point. The same GPT-6 Astra request costs$5.00 per million input tokens in batch and $60.00 on the Ultrafast tier, a twelve-fold spread with no change to the model or the prompt.
OpenAI tiers
| Model | Standard in / out | Batch (0.5×) | Flex (0.5×) | Fast (2×) | Ultrafast (6×) |
|---|---|---|---|---|---|
| GPT-6 Astra | $10.00 / $50.00 | $5.00 / $25.00 | $5.00 / $25.00 | $20.00 / $100.00 | $60.00 / $300.00 |
| GPT-6.1 Sol | $2.00 / $10.00 | $1.00 / $5.00 | $1.00 / $5.00 | $4.00 / $20.00 | — |
| GPT-6 Luna | $0.10 / $0.50 | $0.05 / $0.25 | $0.05 / $0.25 | $0.20 / $1.00 | — |
- Batch: Asynchronous processing. Half the standard price on every model.
- Flex: Slower responses, billed at batch rates.
- Fast: Priority capacity at twice the standard price.
- Ultrafast: GPT-6 Astra only, and only below the long-context threshold.
Anthropic tiers
| Model | Standard in / out | Batch (0.5×) | Fast mode (2×) | US data residency (1.1×) |
|---|---|---|---|---|
| Claude Opus 5.5 | $4.00 / $20.00 | $2.00 / $10.00 | $8.00 / $40.00 | $4.40 / $22.00 |
| Claude Sonnet 5.5 | $2.00 / $10.00 | $1.00 / $5.00 | — | $2.20 / $11.00 |
| Claude Fable 5.1 | $10.00 / $50.00 | $5.00 / $25.00 | — | $11.00 / $55.00 |
- Batch: Asynchronous processing. Stacks with the cache multipliers.
- Fast mode: Research preview, Opus models only, on the first-party API.
- US data residency: inference_geo set to us, on Claude 4.6 and later. Applies to every token category.
Google tiers
| Model | Standard in / out | Batch (0.5×) | Priority (1.8×) |
|---|---|---|---|
| Gemini 3.8 Flash | $0.75 / $3.75 | $0.375 / $1.875 | $1.35 / $6.75 |
| Gemini 3.5 Flash Lite | $0.30 / $2.50 | $0.15 / $1.25 | — |
- Batch: Asynchronous processing.
- Priority: Confirmed for Gemini 3.8 Flash on the pricing page.
DeepSeek tiers
| Model | Standard in / out | Peak hours (2×) |
|---|---|---|
| DeepSeek V4.1 Flash | $0.15 / $0.60 | $0.30 / $1.20 |
| DeepSeek V4 Pro | $0.66 / $1.98 | $1.32 / $3.96 |
- Peak hours: 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday. Catalogue rates are the off-peak ones.
Batch is the only free lunch
Batch halves the price on OpenAI, Anthropic and Google, and the only thing you give up is immediacy. Work that can wait a few hours costs half as much: evaluation runs, bulk classification, embedding backfills, overnight content generation.
OpenAI's Flex tier charges the same as batch while still answering in a single request, for workloads that can accept slower responses but not a queue.
The opposite end is rarely worth it by cost alone. Ultrafast is six times the standard price and exists for one model, so it is a latency purchase, not an economy one. Measure what a second of latency is worth to your product before buying it.
Multipliers stack, so multiply them
Anthropic says its batch discount, cache multipliers and data-residency premium stack. A cached read on the batch tier with US-only inference is 0.1 × 0.5 × 1.1 of the base input rate, not a sum of three separate discounts.
DeepSeek's time-of-day pricing behaves the same way: the peak multiplier applies to the cached rate as well, so a cached read during peak hours costs double a cached read at night.
Our calculator and comparison pages use standard-tier prices, because that is what most requests pay. To price another tier, multiply by the figure in the tables above.
Measure against the model you use
Other comparisons worth running
Compare approaches, not just models
Frequently asked questions
Why does one model have several prices?+
Providers sell the same model at different speeds and priorities. OpenAI has Batch and Flex at half price, Fast at twice the price, and Ultrafast at six times. Anthropic has Batch at half price and a Fast mode at twice the price on Opus models. Google has Batch at half price and a Priority tier at 1.8 times. DeepSeek charges double during peak hours.
How much does OpenAI Ultrafast cost?+
Six times the standard price, and only on GPT-6 Astra below the long-context threshold. That is $60.00 per million input tokens and $300.00 per million output tokens, the highest published text price in the market.
Do tier discounts stack with prompt caching?+
Yes. Anthropic states that its batch discount and cache multipliers stack, and that a US data-residency setting multiplies every token category by 1.1 on top. Combine them by multiplying, not by adding.
When is batch processing worth it?+
Whenever the work can wait. Batch halves the price on OpenAI, Anthropic and Google, and the only cost is latency, usually up to 24 hours. Evaluation runs, bulk classification, embeddings backfills and content generation pipelines all fit.
Which prices does this calculator use?+
The standard tier, because it is the one most requests use. Multiply by the tier figure in the tables above for batch, fast or peak-hour pricing.
Sources
Checked: 2 October 2026. Only tiers published on a provider page are listed.