Claude Opus 4.8

If Opus 4.7 is the top shelf, Opus 4.8 is the newest thing someone put on it. Same family, same role in this pack - the flagship Anthropic reasoning node served through Runware's cloud - but the newest weights, which is why you'd specifically reach past the 4.7 node next to it. In a ComfyUI graph, both are the "send the genuinely hard decision here" node; 4.8 is the one to try first when 4.7's answer wasn't good enough.

The honest framing: this is a frontier model you rent by the token. No weights, no local install, no way to run it yourself - which is exactly why it's in this pack. The anthropic:[email protected] model id never appears in your workflow; you just get the STRING back.

What you set

messages is required (RUNWARE_MESSAGES socket). The surface matches the 4.7 node almost exactly:

• settings.thinkingLevel - off / low / medium / high / max / xhigh / adaptive. The whole reasoning range is here; adaptive lets the model budget its own thinking.
• settings.maxTokens - up to 128000, default 4096. Real reasoning tasks will want more than the default; short decisions should keep it low.
• settings.systemPrompt - role and context for the run.
• images - IMAGE socket for visual input.
• toolChoice - gated, with type/name and advanced_json for tools, inputs.documents, settings.stopSequences.
• settings.cache.scope / settings.cache.ttl - Anthropic prompt caching; the difference between "affordable in a batch" and "why is my bill huge."
• includeUsage - token stats.

Output is text (STRING), which is where the workflow actually consumes the result - feed it into a display node, a decision point, or downstream prompt inputs.

Installing

Part of the Runware/ComfyUI-Runware pack.

ComfyUI Manager: search Runware, install, restart.

Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

Deps are runware-sdk, pillow, soundfile. Runware API key from runware.ai/api-keys (Settings panel or RUNWARE_API_KEY).

Where people get burned

Version churn is the trap. 4.8 being newer doesn't mean it's better for your task - frontier model point releases trade strengths, and if 4.7 already nails your workflow, upgrading is just a higher bill. Test both once on a real problem before you commit a graph to 4.8. And the universal Opus rules apply: don't leave it at xhigh for routine work, set a task-appropriate maxTokens, and cache your system prompt.