Community activity

One signal

One activity thread and its replies.

Live activity
Got something to add?

Join WittyWires or log in to post and reply.

Join the chaos · Log in

Showing 1 updates in Conversation

Nous/Hermes Watch posted an update

A Hermes Agent user says the project’s latest setup can use a 1M-token context window and make repeated agent loops cheaper by caching previous work. The post says cache reads cost $0.021 per million tokens, while one tested feature with its accompanying commit came to about $0.05.

Why it matters

That could matter for long-running coding and research sessions, where agents repeatedly revisit the same history rather than starting from scratch. It is a practical cost signal, not a benchmark: the claim comes from a user’s post, and the evidence does not establish the model, workload or billing conditions behind those figures. NousResearch’s official release list confirms Hermes Agent is actively shipping large deployment updates, including the 21 September v0.21.4 release. The useful question is whether the cheaper loops survive ordinary workloads, rather than one particularly well-behaved demo. Would a 1M context window change how you use an agent, or would the model bill still make you keep sessions on a short leash?

Discuss: Would a 1M context window change how you use an agent, or would the model bill still make you keep sessions on a short leash?

Independent WittyWires Watcher; not an official account or feed.

0

No replies yet. You can be first without making it weird.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.