Structured extraction
Extract company, invoice_number and amount_usd from the input. Return only JSON with those keys. Use null for missing values.
Invoice INV-1042 Northline Studio Total due: USD 218.40
Haiku 5.5 Comparison Tool
Independent third-party toolChat with Claude Haiku 5.5 online or compare models side by side using the same prompts. Test writing, data extraction and support tasks, then review saved answers against your own criteria.
The model
Claude Haiku 5.5
Haiku 5.5 is an Anthropic language model. Whether it can replace another model depends on your inputs, the reasoning setting and what an acceptable answer means for your work.
Test extraction, support replies and other recurring text tasks against a baseline. Keep the instructions identical, then judge each answer against your own criteria. A lower API price alone does not establish better value.
Haiku 5.5 overview and limits- Manufacturer
- Anthropic
- Quality evidence
- Your saved outputs and reviews
- Affiliation
- Independent; not Anthropic or OpenAI
Five models, one task
Model reference
Official API prices below are separate from site credits. Connection limits appear only after verification; no quality ranking is implied.
| Model | Official input / output USD per 1M tokens | Manufacturer limits Context / max output | Verified site limits Context / max output | Site reasoning options |
|---|---|---|---|---|
| Claude Haiku 5.5Anthropic | $0.10 / $0.50Standard tier $0.50 / $2.50From 100,001 input tokens Source · 2026-10-09 | 1,000,000 / 128,000Specs · 2026-10-09 | Awaiting verification | Awaiting verification |
| Claude Sonnet 5.5Anthropic | $2 / $10Standard tier Source · 2026-10-09 | 1,000,000 / 128,000Specs · 2026-10-09 | Awaiting verification | Awaiting verification |
| GPT-6 LunaOpenAI | $0.10 / $0.50Standard tier $0.20 / $0.75From 272,001 input tokens Source · 2026-10-09 | 1,050,000 / 128,000Specs · 2026-10-09 | Awaiting verification | Awaiting verification |
| Claude Opus 5.5Anthropic | $4 / $20Standard tier Source · 2026-10-09 | 1,000,000 / 128,000Specs · 2026-10-09 | Awaiting verification | Awaiting verification |
| GPT-6.1 SolOpenAI | $2 / $10Standard tier $4 / $15From 272,001 input tokens Source · 2026-10-09 | 1,050,000 / 128,000Specs · 2026-10-09 | Awaiting verification | Awaiting verification |
Manufacturer API prices
Official cost calculator
Estimate direct API spending at published manufacturer rates. This is an estimate, not your actual API usage cost or a charge to your site credits.
Anthropic source · 2026-10-09
Anthropic source · 2026-10-09
Baseline minus Haiku (USD)
95% difference. Negative means Haiku costs more.
Equal token counts assumed for both models. Billable output includes reasoning tokens; do not add them twice. Standard real-time pricing, excluding cache discounts, batch discounts and taxes. Long-input tiers apply to the whole call. Rates are dated snapshots, not live quotes.
Input samples
Start with a task you can judge
Support triage
Classify the ticket as billing, technical or account. Return JSON with category and one sentence explaining the evidence. Do not include personal data.
I changed my card yesterday but the renewal still failed. I can log in and all my files are there.
Constrained reply
Write a friendly customer support reply in under 70 words. Acknowledge the issue and ask for the order number. Do not promise a refund or claim to have checked an account.
My order has not arrived, and the tracking link stopped updating three days ago.
Site credits
Pay once. Use as needed.
All packs unlock batch comparisons permanently. Credits do not expire. No subscription or automatic renewal. Tax, if applicable, is shown at checkout.
New anonymous sessions receive 120 trial credits when available. A verified account receives a further 600 credits once. Trial credits cover single comparisons and Chat; a successful pack purchase unlocks batches.
FAQ
Questions before you run
No. Haiku5-5.com is an independent third-party service. Model names and trademarks belong to their respective owners.
Submitted model calls use the input and billable output rates shown on the pricing page. Reasoning tokens are included in billable output. A comparison is settled as a whole row; if one side definitely fails, that row is free. Truncated answers and answers you dislike still incur usage. An explicit Chat summary is a separate paid call; you confirm it before it runs.
A successful credit-pack purchase permanently unlocks batches. Each row uses the same model settings for both outputs, and consumes its own credits. A low balance pauses unstarted rows; top up and resume from private history. CSV uploads have a 5 MiB file limit and a 4 MiB submitted-request limit. Oversized requests are rejected, never silently shortened.
No. You mark results acceptable or unacceptable against your criteria. Only reviewed, complete, successful pairs enter the acceptance denominator. Unreviewed and truncated results stay visible without becoming a quality score.
Records are private to your session or account. Request content is processed by external AI service providers for generation. Deleting a record removes its content; necessary billing evidence is retained without the prompt or answer. See the privacy policy before submitting sensitive data.
Email [email protected] with your order reference and the issue. Each request is reviewed individually. See the refund policy.