Claude Fable 5 already in NeuralSpace: the first model of the Mythos class

Briefly about the main thing (BLUF)

Claude Fable 5 - the first model of the Mythos class from Anthropic - is already working in NeuralSpace. We explain what Mythos means, how much access costs, how to try Fable in chat and in the code section, and who needs it first.

Anthropic released Claude Fable 5 - and this is not another “became 3% better in benchmarks.” For the first time in a long time, they have a whole new class of models: Mythos. It stands ABOVE the Opus, which until now was considered the ceiling. We connected Fable 5 to NeuralSpace on the day we got to its API, I tell you what kind of beast it is and how much it costs.

Mythos - what is this anyway?

Previously, the Claude line was easy to read: Haiku is fast and cheap, Sonnet is the golden mean, Opus is the smartest. Now there is a fourth floor on top. Anthropic has split it into two models: Fable 5 is available to everyone via API, and Mythos 5 is only available to select organizations under a separate agreement. Under the hood, according to Anthropic itself, these are the same model, the difference is in additional restrictions.

From a practical point of view: the context is a million tokens. That's about 10 volumes of War and Peace per request. An entire average project with backend and frontend fits into one window, and the model does not “forget” the start of the conversation after an hour of work.

How much does it cost and where is the catch?

The catch is the price, to be honest. Anthropic's official fee is $10 per million input tokens and $50 per million output tokens. Twice as expensive as Opus 4.8. In NeuralSpace, due to partner access, it turned out to be cheaper than the direct API: 760 tokens per million at the input and 3800 per million at the output.

Sounds abstract? Here are the specifics: a regular question and answer in the chat cost me 0.54 tokens. Long code parsing for a couple of thousand lines - about 10-20 tokens. Daily bonus tokens are enough to test the model without replenishment at all.

How to try in chat

Opening "Chat" section, in the drop-down list of models, select Claude Fable 5 - it is at the very bottom, after Opus 4.8. All. Registration by mail, no VPN required, no foreign card required.

Internet search works - check the box under the model selection. You can also attach pictures and documents: PDF, TXT, DOC.

Fable in the code section

And here is the most interesting thing. IN "Code" section Fable 5 is available as an agent model: you describe the project in words, it writes the files itself, installs dependencies, and raises the site on hosting. Previously, we recommended Opus for complex projects - now there is a more powerful option.

The million-dollar context is truly revealed here: the agent keeps the entire project in his head as a whole, and not in pieces. There are fewer situations of “the model rewrote a function and broke three neighboring ones.”

Who really needs it?

Not everyone, and that's okay.

• Simple questions, translations, posts for social networks - take GPT-5 Mini or Sonnet, it will be much cheaper.
• Intermediate coding - Opus 4.8 is still a great choice, and it's half the price.
• But if you have a large project, long documents, complex architecture, or a task on which other models begin to go around in circles, Fable 5 is noticeably more confident. I fed her a confusing bug where Opus suggested “restart and see,” Fable found the reason from the second message.

By the way, if you are choosing between the flagships of different companies, we have detailed comparison of ChatGPT, Claude and Gemini.

Documentation

For those who want to dig deeper: the official announcement is at Anthropic website, description of API and tariffs - in documentation Claude. The model ID is BLK0, the request format is the same as the other models Claude (Messages API), so the existing code is moved by replacing one line.

Well, if without a code, the model is already waiting in chat NeuralSpace. Daily bonus tokens are awarded every day, enough to get acquainted.

Frequently asked questions (FAQ)

Question: How to get the best result from a neural network?
Answer: Use detailed prompts (descriptions) in English, set the style and details of the scene.

Question: Can these materials be used for commercial purposes?
Answer: Yes, the generated content is entirely yours.