Claude Haiku 5.5
- Input / 1M tokens
- $0.10
- Output / 1M tokens
- $0.50
API models · Side by side
Claude Haiku 5.5 vs GPT-6 Luna
Haiku 5.5 and GPT-6 Luna tie on base input and output rates. That makes the shape of your requests more useful than the headline price when choosing between them.
GPT-6 Luna
- Input / 1M tokens
- $0.10
- Output / 1M tokens
- $0.50
The short answer
Short requests cost the same for an equal token count. Luna has slightly more context capacity, while the two models enter higher pricing tiers at different prompt lengths.
Compare the details
Compare context, supported inputs and usage costs. Highlighted rows show a difference.
| Feature | Claude Haiku 5.5 | GPT-6 Luna |
|---|---|---|
| Input formats | file, image, text | file, image, text |
| Output formats | text | text |
| Context / position limit | 1,000,000 | 1,050,000 |
| Maximum output tokens | 128,000 | 128,000 |
| Developer features | Reasoning output, Completion limit, Output limit, Reasoning controls, Thinking effort, Response format, Stop sequences, Structured output, Tool selection, Tool calling, Response length | Reasoning output, Completion limit, Output limit, Reasoning controls, Thinking effort, Response format, Seed, Structured output, Tool selection, Tool calling, Response length |
| Cached input / 1M tokens | $0.01 | $0.01 |
| Cache write / 1M tokens | $0.13 | $0.13 |
| USD input / 1M tokens | $0.10 | $0.10 |
| USD output / 1M tokens | $0.50 | $0.50 |
| Model identifier | anthropic/claude-haiku-5.5 | openai/gpt-6-luna |
API pricing
Prices are in USD per million tokens. Input is what you send; output is what the model generates. Claude and ChatGPT subscriptions are billed separately.
Claude Haiku 5.5: long-prompt rates
- Long-prompt threshold: 100,000 prompt tokens. Input: $0.50 / 1M. Output: $2.50 / 1M.
GPT-6 Luna: long-prompt rates
- Long-prompt threshold: 272,000 prompt tokens. Input: $0.20 / 1M. Output: $0.75 / 1M.
Tools, images and additional reasoning can add charges. The examples below cover input and output tokens only.
Compare local modelsWhat those prices mean in practice
Each example uses the same token budget for both models. These are calculations, not measured workloads. Cache discounts, media, tools and extra reasoning are excluded.
| Workload | Claude Haiku 5.5 | GPT-6 Luna |
|---|---|---|
| 1,000 short requestsPer request: 2,000 input + 500 output tokens | $0.45 | $0.45 |
| 100 document summariesPer request: 20,000 input + 1,000 output tokens | $0.25 | $0.25 |
| 10 long-document requestsPer request: 300,000 input + 2,000 output tokens | $1.55 | $0.61 |
Estimate = requests × (input tokens × input price + output tokens × output price) ÷ 1,000,000. Long-prompt rates are applied where the pricing data provides them. Real requests can use different amounts of output.
Beyond the price tag
Why a base-price tie is not a monthly-cost tie
A support assistant can produce a short answer with one model and a much longer answer with another. The rates may match while the bill differs. Budget the actual input, returned text and any billable reasoning for the entire conversation. Include failed attempts that are billed. Use the short-request example as a starting calculation, then replace its assumptions with the requests your application sends.
The long-document decision
Haiku and Luna have different long-prompt thresholds and higher-tier rates. A large reference pack can therefore change this comparison even when ordinary messages cost the same. Conversation history, instructions and attached material contribute to prompt size. Keep room for the reply as well. If the application needs only a few passages from a report, retrieving those passages can be more economical than sending the whole report every time.
Controls that affect an integration
Both support tool use and structured responses in the current data. Luna also exposes a seed control here, while Haiku includes a stop control. Treat these as integration requirements, not quality scores. A seed is not a promise of identical answers under every condition, and a stop sequence is not a replacement for validating returned JSON. Check which controls your current code actually relies on.
Choosing for repetitive work
For extraction, tagging and short summaries, make the expected output small and explicit. Keep a few difficult examples alongside ordinary ones so missing fields and ambiguous requests are represented. If you already have a dependable integration with either vendor, a price tie alone gives little reason to move. The stronger reason is a concrete improvement in your application’s results, long-document budget or required controls.
Common questions
Which is cheaper for short requests?
At the current base rates, they tie when the billable input and output token counts are equal. Output length, reasoning and optional services can change an actual request total.
Can either process recordings directly?
The compared endpoints accept text, images and files. Audio and video are not input formats in this pair; a recording would require an appropriate conversion service or a different model.