Compare Claude Sonnet 5.5 vs Haiku 5.5 OpenRouter price and context
Both ids list a 1,000,000 context. Input is $2.00 vs $0.10 and output is $10.00 vs $0.50 per million tokens. A lower price is not a worse model.
To compare Claude Sonnet 5.5 and Claude Haiku 5.5, start with context length and list price on the same day’s catalog. Cheaper only means a lower unit price on that row.
The query is Compare Claude Sonnet 5.5 vs Haiku 5.5. Figures come from the OpenRouter model list fetched on 2026-11-02 (openrouter.ai/api/v1/models) and from Hugging Face model search. This page does not invent eval scores, and it does not treat the lower price as a worse model.
Catalog snapshot on 2026-11-02
| Field | Claude Sonnet 5.5 | Claude Haiku 5.5 |
|---|---|---|
| OpenRouter id | anthropic/claude-sonnet-5.5 | anthropic/claude-haiku-5.5 |
| context_length | 1000000 | 1000000 |
| Input / 1,000,000 tokens | $2.00 | $0.10 |
| Output / 1,000,000 tokens | $10.00 | $0.50 |
| List-price multiple (input / output) | Input is 20× Haiku 5.5; output is also 20× | Input is 1/20 of Sonnet 5.5; output is also 1/20 |
| Cache read / 1,000,000 tokens | $0.10 | $0.01 |
| Cache write / 1,000,000 tokens | $2.50 | $0.125 |
| 1-hour cache write / 1,000,000 tokens | $4.00 | $0.20 |
| Per-token list price (input / output) | 0.000002 / 0.00001 | 0.0000001 / 0.0000005 |
| web_search raw field (not converted) | 0.01 | 0.01 |
| pricing.overrides | Not present on this response | At min_prompt_tokens 100000: input $0.50, output $2.50 per million |
| OpenRouter created | 1790618686 · 2026-09-28 | 1791397883 · 2026-10-07 |
| canonical_slug | anthropic/claude-sonnet-5.5-20260928 | anthropic/claude-haiku-5.5-20261007 |
| Input → output | text, image, file → text | text, image, file → text |
| top_provider.max_completion_tokens | 128000 | 128000 |
| hugging_face_id | null | null |
| Official Hugging Face weights | None. author=anthropic returned 0 | None. author=anthropic returned 0 |
What matches is the context length. What differs is the list price.
Turn dollars per token into dollars per million
prompt, completion, input_cache_read, input_cache_write, and input_cache_write_1h are US dollars per token. Multiply by 1,000,000. Sonnet 5.5 is then $2.00 input, $10.00 output, $0.10 cache read, $2.50 cache write, and $4.00 for the 1-hour cache write. Haiku 5.5 is $0.10 input, $0.50 output, $0.01 cache read, $0.125 cache write, and $0.20 for the 1-hour cache write. web_search is 0.01 on both rows; leave that field as published and do not multiply it by one million.
per_million = per_token * 1000000
sonnet prompt 0.000002 -> $2.00
sonnet completion 0.00001 -> $10.00
haiku prompt 0.0000001 -> $0.10
haiku completion 0.0000005 -> $0.50The long-prompt override is still a list price
When Haiku 5.5 prompt tokens reach min_prompt_tokens 100000, prompt is 0.0000005 and completion is 0.0000025, which is $0.50 input and $2.50 output per million. In that same override, cache read is $0.05, cache write is $0.625, and the 1-hour cache write is $1.00. Sonnet 5.5 pricing in this fetch has no overrides. Even this higher Haiku tier stays below Sonnet 5.5’s base list price. That is still a price difference.
Catalog blurbs are not scores
What the Sonnet 5.5 blurb says
OpenRouter describes it as a Sonnet-class model for well-scoped everyday work, and says it succeeds Claude Sonnet 5. That sentence points at a different id, anthropic/claude-sonnet-5, which is not a row in this table.
What the Haiku 5.5 blurb says
OpenRouter describes it as a small, fast model for high-volume, cost-sensitive work such as summarization, subagents, and browser use. That is a blurb. A lower unit price does not make the model worse.
Hugging Face has no official weights
On 2026-11-02, searches for anthropic/claude-sonnet-5.5 and anthropic/claude-haiku-5.5 each returned 0 models. author=anthropic also returned 0. A normal search does not show official Anthropic weights. Name-similar community repos (for example diablomike20/Claude-Sonnet-5.5, and Haiku 5.5 repos whose names include Qwen) are not official weights for these two OpenRouter ids. This table does not use their download counts.
Check the same fields yourself
- 01Lock the two ids with no suffix
Find
anthropic/claude-sonnet-5.5andanthropic/claude-haiku-5.5on the OpenRouter model list. Do not read them asanthropic/claude-sonnet-5, and do not read a:batchid as the same row. - 02Convert only per-token prices
Multiply prompt, completion, and the cache fields by 1,000,000. Leave web_search at
0.01. - 03Read the override separately
Haiku 5.5 has an override at min_prompt_tokens 100000. Sonnet 5.5 does not in this fetch. The override is still a list price.
- 04Use the Hub only to see if an official repo exists
author=anthropic returned 0, so write that official weights are absent. Do not add community download counts to these two rows.
This table is a model catalog, not an oakmeet feature list. To put the price table on one page with a colleague, open a temporary room on oakmeet (browser only, no account). See the guide.
Does the lower list price mean Haiku 5.5 is worse?
No. $0.10 vs $2.00 and $0.50 vs $10.00 are unit prices on the 2026-11-02 list. This page has no quality eval, and it does not treat cheaper as worse.
If context matches, what is left to compare?
On this fetch both ids show context_length 1000000, the same text, image, and file input to text output, and max_completion_tokens 128000. Those catalog fields match. The list price does not.
Does Hugging Face host official weights?
Not in this refetch. Exact id search returned 0, and author=anthropic returned 0. hugging_face_id is null on both OpenRouter rows. Community lookalikes are not official weights.
Is anthropic/claude-sonnet-5 on this table?
No. It is a different id, canonical slug anthropic/claude-sonnet-5-20260630, created on 2026-06-30. The :batch ids for Sonnet 5.5 and Haiku 5.5 are different rows as well.