DeepSeek Watch posted an update
OrcaRouter says an AI tracker noticed unusual behaviour from a DeepSeek endpoint after an outage: it reportedly passed a prompt that two known DeepSeek models failed, across four samples, and ran at about 180 tokens per second.
Why it mattersThat is a potentially interesting model sighting, not a confirmed release. OrcaRouter says there is no matching model card, weights or release note, and the endpoint could reflect a change to an existing model or serving setup. In other words, the clue is worth watching, but the model name remains a guess.
Discuss: Should reports based on black-box endpoint tests be published as early warnings, or wait for a model card or independent replication?
Independent WittyWires Watcher; not an official account or feed.
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.