Anthropic

Anthropic's Claude joins Brut as a planning model — text only. It generates no images or video; it's there to be your director: writing the concept, look and shot plan on the Director node, and powering the Refine node.

#Claude Sonnet 5.5

Sonnet 5.5 is Anthropic's newest Sonnet: faster and sharper than Sonnet 5, at the same price.

• Marcel's default brain. With an Anthropic key, Marcel's Auto brain is now Sonnet 5.5 (it was Sonnet 5). A brain you picked yourself in ⚙ Settings → Marcel stays picked.
• In every model picker: the Director, the crew agents, Text→Text, Refine and Marcel. It runs on your Anthropic key, or through OpenRouter at the same rates.
• Price: $2 in / $10 out per million tokens, the same as Sonnet 5. Marcel's long system context is cached, so after the first message of a session each turn costs cents.
• Reasoning goes from low to max (default high). Thinking off means no thinking before acting on this model, and works at low, medium or high effort only. Details under Reasoning dial & thinking switch.
• Marcel keeps talking while he works. Marcel keeps writing short notes between his actions, so you can follow what he does.

#Claude Haiku 5.5

Haiku 5.5 is Anthropic's new small model: a tenth of Haiku 4.5's price, a 1M-token context, and the same two knobs as the big Claudes.

• The narrator's new default. With an Anthropic key, Marcel's Auto narrator lane (failure diagnoses, run commentary, recaps, the quiet taste passes) now runs on Haiku 5.5. A narrator you picked yourself stays picked. It is also in every text picker: the Director, the crew agents, Text→Text, Refine and Marcel's brain.
• Price: $0.10 in / $0.50 out per million tokens for prompts up to 100,000 tokens. Over 100,000 tokens a prompt pays five times that ($0.50 / $2.50): the one current Claude priced by prompt length. Brut's estimates account for it. A long chat on Marcel's full knowledge can pass the line; the narrator's few hundred tokens never do.
• Reasoning goes from low to max; thinking is adaptive by default and switches off cleanly (a plain off, unlike Sonnet 5.5's "no thinking before acting").
• Runs on your Anthropic key, or through OpenRouter at the same base rates.

#Get a key

Create a key in the Anthropic Console and paste it into ⚙ Settings → Anthropic (Claude). Brut checks it when you save.

#How it's used

Anywhere Brut asks "which LLM should plan this?" — the Director (and the legacy crew agents), Text→Text and the Refine presets — you can pick Claude, Gemini or GPT. Claude reads your stills and writes the plan.

#Reasoning dial & thinking switch

Claude Sonnet 5.5 (Anthropic's newest Sonnet), Sonnet 5, Opus 4.8, Opus 5 and Opus 5.5 carry the knobs Brut exposes wherever these models run — the Text→Text node, the cinema agents, and Marcel's brain lane:

• Reasoning effort (low to max): how deeply the model reasons before answering. Deeper = better on hard treatments, more output tokens billed. Claude's own default is high on every model but Opus 5.5, whose default is medium: on 5.5, set the dial when you want the deep end.
• Thinking switch (Marcel's brain, ⚙ Settings): Claude Sonnet 5.5, Sonnet 5, Opus 4.8 and Opus 5 think adaptively by default, and those thinking tokens bill as output tokens — so if Marcel feels slow or a chat turn costs more than expected, switching thinking off is the lever: faster, cheaper, a little shallower. The switch needs your own Anthropic key; it isn't available through OpenRouter. Opus 5.5 has no switch: its thinking is always on (Anthropic rejects "off"), so the row reads not supported and the cost lever there is a low effort. On Sonnet 5.5, off means "no thinking before acting": Anthropic's lowest setting on that model still lets it write short progress notes between actions, and it works at low, medium or high effort only (with the dial at xhigh or max, Brut asks you to lower it or turn thinking back on).

Haiku 5.5 carries both knobs too (thinking adaptive by default, a plain off). Haiku 4.5 supports neither.

#Which Claude for what

• Sonnet 5.5 ($2 / $10 per million tokens in / out): Marcel's default brain and the everyday planner, faster and sharper than Sonnet 5 at the same price.
• Sonnet 5 ($2 / $10): the previous Sonnet, still listed for graphs and lanes that picked it.
• Opus 5.5 ($4 / $20): the powerhouse brain, Opus-tier reasoning below Opus 5's price. On this model, repeated context in long Marcel sessions costs half the usual.
• Opus 5 and Opus 4.8 ($5 / $25): still listed for graphs and lanes that picked them.
• Haiku 5.5 ($0.10 / $0.50 up to 100K tokens of prompt, $0.50 / $2.50 above): the narrator lane, and the cheapest Claude for Refine and Multi-Shot plans.
• Haiku 4.5 ($1 / $5): the previous narrator, still listed for lanes that picked it.

List prices as of October 2026; the spend panel shows what you were actually billed.

#Notes

• A plan that comes back cut off is never passed on half-written: you get the full plan or a clear error.

For available model names and pricing, see Anthropic's console; Brut's catalog lists the ones it exposes.