2 points | by GodelNumbering 7 hours ago ago
3 comments
I intended to post this exact link (https://dirac.run/labs-market-share#daily-token-share-by-lab) but seems like HN removes the hash part
All the slop has to come from somewhere in the end.
This might be the very exit of the central sewage.
DeepSeek V4.1 Flash 1017, for a set of document extraction tasks we use, is the first model to step change move the cost per task down.
Serving via Deepinfra Flex at $0.0784 / $0.2352 / $0.0024 per M/tok is the least expensive cloud model we have ever benched by a wide margin.
This price / performance is probably the reason for the spike.
I intended to post this exact link (https://dirac.run/labs-market-share#daily-token-share-by-lab) but seems like HN removes the hash part
All the slop has to come from somewhere in the end.
This might be the very exit of the central sewage.
DeepSeek V4.1 Flash 1017, for a set of document extraction tasks we use, is the first model to step change move the cost per task down.
Serving via Deepinfra Flex at $0.0784 / $0.2352 / $0.0024 per M/tok is the least expensive cloud model we have ever benched by a wide margin.
This price / performance is probably the reason for the spike.