Long-context pricing calculator
Some models get dramatically more expensive once a prompt passes a size threshold — and the higher rate applies to the whole request, not just the extra tokens. Pick a tier, set your prompt length, and see where the steps are.
Compare
Claude Haiku 5.5
$0.063
per request · $6,250 a month · over the 100K step
GPT-6 Luna
$0.013
per request · $1,250 a month
At 120K tokens, GPT-6 Luna is cheaper: $0.013 per request against $0.063 — Claude Haiku 5.5 costs 5.0× as much. This prompt is over 100K, so Anthropic bills any Claude Haiku 5.5 request over 100K input tokens at 5× every rate — for the whole request, not just the tokens past the line.
Cost per request by prompt length
Either side of the steps
Cost per request at 1K output tokens. One token either side of a threshold is where the bill jumps.
| Prompt | Claude Haiku 5.5 | GPT-6 Luna | Cheaper |
|---|---|---|---|
| 10K | $0.0015 | $0.0015 | Same |
| 50K | $0.0055 | $0.0055 | Same |
| 100K | $0.011 | $0.011 | Same |
| 100K + 1 | $0.053 | $0.011 | GPT-6 Luna, 5.0× |
| 200K | $0.102 | $0.021 | GPT-6 Luna, 5.0× |
| 272K | $0.139 | $0.028 | GPT-6 Luna, 5.0× |
| 272K + 1 | $0.139 | $0.055 | GPT-6 Luna, 2.5× |
| 500K | $0.253 | $0.101 | GPT-6 Luna, 2.5× |
| 1M | $0.502 | $0.201 | GPT-6 Luna, 2.5× |
Try this: on the Small tier, drag the prompt from 90K to 110K. Claude Haiku 5.5 and GPT-6 Luna have identical prices up to 100K tokens. One token later, Haiku 5.5 costs five times as much — and it stays the dearer of the two all the way to 1M, even after Luna's own step at 272K.
Which models charge more for long prompts?
- Claude Haiku 5.5: a prompt over 100,000 tokens is billed at 5× every rate — input, cache reads, cache writes and output. The prompt length counts cached tokens too.
- GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna: a request with more than 272K input tokens is billed at 2× input and cache rates and 1.5× output.
- Claude Fable 5.1, Opus 5.5 and Sonnet 5.5: one rate across the full 1M-token window. A 900K-token request costs the same per token as a 9K one.
Why the step matters more than the rate
A long-context surcharge isn't charged on the tokens past the line — it reprices the whole request. A RAG pipeline that usually sends 80K tokens of documents is fine on Haiku 5.5; on the day a query retrieves a few extra documents and lands at 105K, that request costs five times as much. If your prompt sizes vary around a threshold, your bill becomes bimodal: forecast it from the distribution of prompt sizes, not the average, and consider capping retrieved context just under the line.
How it's calculated
- Cost = cached prompt × cache-read rate + uncached prompt × input rate + output × output rate, with the long-context multiples applied to every term when the prompt is over the threshold.
- Cached prompt tokens assume the cache was already written by an earlier request. For multi-turn sessions where the cache is written as you go, use the agent loop cost calculator.
- Excludes batch discounts (50% on both providers), fast and priority tiers, and data-residency premiums. Tokenizers differ between providers, so the same text isn't the same number of tokens on each.
- GPT-6 Luna: OpenAI doesn't publish a context window for it, so it's modelled without one.
Prices from Anthropic's and OpenAI's own pricing documentation, checked 17 October 2026.
Related
- Claude Haiku 5.5 vs GPT-6 Luna — same price, until the prompt gets long.
- Agent loop cost calculator — what a whole multi-turn session costs.
- LLM API cost estimator — monthly spend at your volume.