Claude Haiku 5.5 for Chemical Companies: Strong for Its Price, Weak on Documents
Anthropic's Claude Haiku 5.5 rates level with Gemini 4 Argon and GPT-6.1 Sol on real professional work, at $0.21 an index task. On reading whole documents most of the field beats it, so a chemical distributor or producer should test it job by job.
Published
First published.
Claude Haiku 5.5 at a glance
What Claude Haiku 5.5 is.
Anthropic released Claude Haiku 5.5 on 7 October 2026. Anthropic says it is “designed for high-volume, cost-sensitive tasks”, such as summaries, database queries, and classification requests. It reads text and images and costs $0.10 for every million tokens you send and $0.50 for every million it writes, for prompts up to 100,000 tokens. Above that it costs $0.50 and $2.50.
Three of the guide's tests have scored it so far. The scores on this page are at its highest thinking setting, the one each test ran.
Where Claude Haiku 5.5 lands, job by job.
Each line is one test, and each dot is one model it scored, from lowest at the left to highest at the right. The large dot is Haiku 5.5. It sits in the top half on two jobs and well down the field on the third.
Finished professional work
Artificial Analysis, 25 models ranked on real professional tasks
Claude Haiku 5.511th of 25
The test can't separate it from Gemini 4 Argon, GLM-5.3-Flash, or GPT-6.1 Sol, inside its margin of error.
Artificial Analysis: GDPval-AANo date stated
Regulatory research
vals.ai Legal Research, questions answered fully right
Claude Haiku 5.550%$1.44 a question
26th of 76, at 50%. The guide's best-value pick, GLM-5.3 Flash, an open model, scores 57.58% for $0.099 a question.
vals.ai Legal Research, regulatory questions12 Oct 2026
Reading whole documents
ParseBench, a whole-document reading score out of 100
Claude Haiku 5.557.40.14 cents a page
96th of 147. The test is run by a company that sells a parser it ranks, and that parser leads.
ParseBench leaderboard9 Oct 2026
Where Claude Haiku 5.5 falls short.
A small model is cheap for a reason in some places. These are the ones the tests show.
Low on Reading Documents
ParseBench, run by a company that sells a rival tool, puts it 96th of 147 on whole documents and 106th on reading data from charts, at 10.13 out of 100. Test it on your own TDSs, SDSs, and CoAs before it reads them.
ParseBench leaderboard9 Oct 2026
Not the Cheapest Way to Research Rules
It got 50% of US regulatory questions fully right at $1.44 a question. GLM-5.3 Flash, an open model, got 57.58% at $0.099. The test is US law, not REACH or GHS.
vals.ai Legal Research, regulatory questions12 Oct 2026
Cheap Tokens, Many of Them
Its tokens cost a twentieth of Claude Sonnet 5.5's. But on the index it wrote 440 million tokens, where the median model in its class wrote 95 million, so one index task still cost $0.21.
Artificial Analysis: Claude Haiku 5.5No date stated
Its Own Jobs Are Untested
Anthropic built it for summaries, database queries, and classification. No public test the guide uses has measured that work yet, so sorting your RFQs and order emails is a job to test.
Anthropic: Claude Haiku 5.57 Oct 2026
Questions about Claude Haiku 5.5.
Short answers, from the same sources as the rest of this page.
What is Claude Haiku 5.5?
A small model from Anthropic, released on 7 October 2026 for high-volume, cost-sensitive work such as summaries and classification. It reads text and images.
How much does Claude Haiku 5.5 cost?
$0.10 for every million tokens sent and $0.50 for every million written, for prompts up to 100,000 tokens, and $0.50 and $2.50 above that. One task on the Artificial Analysis Intelligence Index cost $0.21.
Claude Haiku 5.5 or GPT-6.1 Sol?
On real professional work the test can't separate them: Haiku ranks 11th of 25 and Sol 13th. An index task cost $0.21 on Haiku and $0.72 on Sol. On regulatory research Sol got 53.03% at $2.76 a question, and Haiku 50% at $1.44.
Can Claude Haiku 5.5 read supplier documents?
It can read them, but on ParseBench it ranks 96th of 147 on whole documents. TeleOCR, a free model, leads the independent document test, so compare both on your own TDSs, SDSs, and CoAs.
Sources.
- Artificial Analysis: models and the Intelligence Index Artificial Analysis, an independent testing company. Each model's record gives its index score, its cost per index task, and its professional work rating.States no date · Vailent checked 12 Oct 2026
- Artificial Analysis: GDPval-AA Artificial Analysis, an independent testing company. It rates models on real professional tasks and publishes a 95% interval for each.States no date · Vailent checked 12 Oct 2026
- Artificial Analysis: Claude Haiku 5.5 Artificial Analysis, an independent testing company. It ranks the model against the 175 models in its price class.States no date · Vailent checked 12 Oct 2026
- vals.ai Legal Research, regulatory questions vals.ai, an independent testing company. Lawyers wrote and checked the questions; an answer counts only if it meets every point on the lawyers' list.Published 12 Oct 2026 · Vailent checked 12 Oct 2026
- ParseBench leaderboard LlamaIndex, which sells LlamaParse, a product it ranks.Published 9 Oct 2026 · Vailent checked 12 Oct 2026
- Anthropic: Claude Haiku 5.5 The maker's own page.Published 7 Oct 2026 · Vailent checked 12 Oct 2026