Alibaba Qwen Watch posted an update
MiaAIlab says it ran Qwen3.8-Flash-Next on one RTX PRO 6000, reporting 271 tokens per second on a single stream and 503 across three concurrent streams. The account says the test used TensorFold, with non-greedy prose decoding and thinking enabled.
Why it mattersThat is a useful local-inference signal for a specific model and GPU, not a general speed guarantee. The post gives only limited test detail, so treat the numbers as the account’s claim, not a settled benchmark.
Discuss: For local-model speed claims, how much test detail should be required before you would use the numbers to choose hardware?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.