Anthropic released Claude Haiku 5.5 on October 7, and it replaced Haiku 4.5 in Telematic the same day. Benchmarks, where it runs, why your price is unchanged.
On October 7, 2026, Anthropic released Claude Haiku 5.5, the smallest and fastest model in the Claude family. It went live in Telematic the same day and replaced Haiku 4.5. Your price has not changed: Haiku 5.5 costs the same as Haiku 4.5, 1 token for a post. Below: what this model is, what the benchmarks show, where it runs in Telematic and when another model is the better choice.
A short video on the same topic, under a minute with Kubit: Claude Haiku 5.5 is now in Telematic (in Russian).
The essentials and the limitations first
- The benchmark figures come from Anthropic. We have not re-tested them on our own tasks. We will collect a comparison on Telematic posts and scripts from real projects and publish it separately.
- Haiku 5.5 is not the strongest model on the list. According to Anthropic itself, Claude Sonnet 5.5 is ahead on complex multi-step tasks. It is also available in Telematic and costs more: a post for one platform is 2 tokens.
- GPT-6 Luna writes texts by default. Haiku 5.5 is selected in the model list (see "How to turn it on" below). If you previously had Haiku 4.5 selected, your choice has already switched to Haiku 5.5 on its own.
- Video and AI images in Telematic unlock after your first top-up of at least 100 tokens (about $1). Text operations on Haiku 5.5 are available right away, including on welcome tokens.
In short: what changed for you
| What | Before | Now |
|---|
| The cheapest Claude model | Claude Haiku 4.5 | Claude Haiku 5.5 |
| Post for one platform | 1 token | 1 token |
| Your saved choice of Haiku 4.5 | β | switched to Haiku 5.5 automatically |
| Default "AI graphics" model in video editing | Haiku 4.5 | Haiku 5.5 |
1 token = $0.01 = β¬0.01.
What Claude Haiku 5.5 is
Haiku is Anthropic's line of small, fast models. They are used where requests are numerous and speed and price matter: short texts, summaries, classification, support replies. The larger models in the same family are Sonnet 5.5 and Opus 5.5.
What Anthropic says about the new version:
- Speed. Anthropic's fastest model at standard speed.
- Reasoning depth. The first Haiku model with an effort setting: you can ask it to think longer on a hard task or answer quickly on a simple one. Levels range from "low" to "max".
- Tasks. Summarizing and condensing text, data queries, classification, acting as an assistant alongside larger models, real-time customer support, working in a browser.
- New tokenizer. The same text is split into slightly more tokens. This does not affect your price in Telematic: you pay per operation (a post, a rewrite), not per model token.
- Safety. Anthropic reports noticeably fewer cases of undesirable behavior compared with Haiku 4.5.
- Availability. On all Anthropic channels, including the Amazon Web Services, Google Cloud and Microsoft Azure clouds.
What the benchmarks show
Results from Anthropic's launch page for the model (checked October 7, 2026). The GPT-6 Luna column uses data from the same page; this model is also available in Telematic at the same price as GPT-6 Luna.
| Benchmark | What it tests | Haiku 4.5 | Haiku 5.5 | GPT-6 Luna | Sonnet 5.5 |
|---|
| OSWorld 2.1 (offline set) | using a computer: windows, programs, files | 15.7% | 72.4% | 48.9% | 83.9% |
| Humanity's Last Exam, no tools | very hard questions across the sciences | 10.2% | 45.9% | β | 56.9% |
| Humanity's Last Exam, with tools | the same, with search and code | 18.7% | 57.4% | β | 64.5% |
| Chartography, no tools | reading charts and diagrams | 6.4% | 46.4% | 29.1% | 61.6% |
| GDPval-AA v2.1, Elo rating | office tasks from real professions | 735 | 1620 | 1437 | 1840 |
| Terminal-Bench 4.0 | hard command-line tasks | 0.0% | 39.2% | 16.4% | 70.6% |
How to read this. On every benchmark in the table, Haiku 5.5 is far ahead of Haiku 4.5, and on most of them it is also ahead of GPT-6 Luna. Sonnet 5.5 leads everywhere; the gap is largest on Terminal-Bench, which tests long multi-step coding tasks.
Companies that tested the model before launch report similar results. For example, Box saw its score rise by 11 points compared with Haiku 4.5 at roughly half the latency, and Asana saw task completion latency drop by more than 30%. These are their own tasks; the results do not carry over directly to yours.
What it costs
In Telematic. Haiku 5.5 costs the same as GPT-6 Luna and Gemini 3.5 Flash Lite: almost everything is 1 token. Operations cost the same as they did on Haiku 4.5:
| Operation | Tokens on Haiku 5.5 |
|---|
| Post for one platform (creating or regenerating) | 1 |
| "Rephrase Text" when creating content | 1 |
| "Write with AI" | 1 |
| Advisor ideas, draft from an idea | 1 |
| AI graphics over video in editing | 1 for planning + 1 for each started 10 seconds |
A full post as an example. Collecting an article from RSS (1) + a Telegram post on Haiku 5.5 (1) + scheduled publishing (1) = 3 tokens, about $0.03. On Sonnet 5.5, which also does the collecting, the same post costs 8 tokens: collection (5) + the post (2) + publishing (1).
From Anthropic (for those who connect the model through the API themselves). Per million tokens: input $0.10, output $0.50 for requests up to 100 thousand tokens; above 100 thousand, $0.50 and $2.50. Haiku 4.5 cost $1 and $5. By Anthropic's estimate, running it costs on average about 75% less than Haiku 4.5.
Where Haiku 5.5 runs in Telematic
- Post and content texts. Everywhere there is an "AI model" list: a post for each platform, "Shorten" and "Adapt", rewriting and "Write with AI" when creating content, the AI generator as a source, advisor ideas. Your choice is remembered for the next text operations.
- Project AI settings. You can choose the model for the whole project, and new texts will then use it.
- AI graphics in video editing. Numbers, key points and headlines over the video: in the "Graphics model" switch, the Haiku option is the default and now runs on 5.5.
- Video scripts. You can choose Haiku 5.5 as the script model. By default, scripts are written by Sonnet 5.5: a long, coherent text with a hook is exactly the kind of task where the larger model is stronger. The "Reasoning depth" setting ("Fast", "Balanced", "Deep") also works for Haiku 5.5.
To keep Haiku 5.5 fast, Telematic sends it a low reasoning depth for regular text operations. It thinks more deeply when you choose "Balanced" or "Deep" for a script yourself.
How to turn it on
- Open a text operation, for example "Post for platform", or "AI Settings" on the project page.
- In the "AI model" block, in the Anthropic group, choose "Claude Haiku 5.5". The operation's price in tokens is shown next to it.
- Run the operation. The model will be remembered for the next texts.
If you already had Haiku 4.5 selected, there is nothing to do: your choice switched on its own.
No account yet? Sign up at telematic.pro: you get 50 welcome tokens on sign-up, enough to compare posts on Haiku 5.5 and other models. All prices are on the pricing page.
Which model to choose
| Task | What to use | Why |
|---|
| Daily posts from sources, rewriting, short texts | Haiku 5.5 or GPT-6 Luna | fast and 1 token per post; compare both on your channel over 3β5 posts |
| Captions, graphics with numbers in video editing | Haiku 5.5 (the default) | short text, speed matters |
| Video script, long article, complex topic | Sonnet 5.5 | stronger on complex multi-step tasks in Anthropic's benchmarks |
The most reliable way to choose is to make one post on each model from your own source and compare. Regenerating a post for a platform costs 1 token on Haiku 5.5 and GPT-6 Luna.
Frequently asked questions
Do I need to switch anything if I used Haiku 4.5?
No. Your saved choice of Haiku 4.5 moved to Haiku 5.5 automatically, at the same price.
Have posts become more expensive?
No. Haiku 5.5 costs the same as Haiku 4.5. A post for one platform costs 1 token.
Can I go back to Haiku 4.5?
No, it is no longer on the list. If Haiku 5.5's output does not suit you, try GPT-6 Luna or Gemini 3.5 Flash Lite at the same price, or Sonnet 5.5, which costs 2 tokens per post.
Will the texts get better?
In Anthropic's benchmarks the model is noticeably stronger than its predecessor. Only a trial will show the difference on your texts: tell us what you got, and we will collect examples and comparisons in a separate article.
Sources: Anthropic's page on Claude Haiku 5.5 (checked October 7, 2026); Telematic price list in the help center article "Tokens and payment: the full price list". 1 token = $0.01 = β¬0.01.