PLATFORMS Signal 314 2 feeds carried it
DeepSeek V4 Flash now runs updated weights on AI Gateway
Engineers already routing requests to deepseek/deepseek-v4-flash automatically receive the updated weights, eliminating migration effort. The new weights show a 25.8-point jump on Terminal-Bench (56.9 to 82.7), meaning agentic workflows like coding tasks should see materially better results. DeepSeek is currently the only provider serving these weights, so teams requiring alternative providers or Zero Data Retention arrangements will need to wait until next week.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
The updated weights score 82.7 on Terminal-Bench, up from 56.9 in the April preview, reflecting substantially enhanced agentic capabilities.
Requests to the deepseek/deepseek-v4-flash model ID automatically pick up the new weights with no code changes required.
DeepSeek is the sole provider of the updated weights today; other providers, including Zero Data Retention options, are expected next week.
THE CLUSTER
↗