DeepSeek releases its model weights as open source; AI coins make an initial move
DeepSeek V4.1 Flash has switched to version 1017, with its weights released directly under the MIT license. DeepSWE v1.1’s software engineering pass rate rose from 74.2% to 77.0%, surpassing Opus 5’s 74.0%, at a cost of 1.3 yuan per problem. The Search and Fetch APIs have also opened up, priced at 5 yuan per 1,000 requests.
With inference costs plunging to this level, the cost of calls for on-chain AI agents is collapsing too, giving narratives around agent orchestration some room for short-term speculation. They benefit from surplus inference compute, but demand-side growth is unlikely to show up in the near term. The cost structure of subnet mining is taking a direct hit, and spot sell orders for leading subnet tokens have thinned noticeably. This is an event-driven spike; with heavy overhead supply, a pullback and consolidation after the surge seems more likely.
Are you chasing the first green candle here, or waiting for a retest of the previous low before getting in?
$FET $RENDER $TAO
DeepSeek V4.1 Flash has switched to version 1017, with its weights released directly under the MIT license. DeepSWE v1.1’s software engineering pass rate rose from 74.2% to 77.0%, surpassing Opus 5’s 74.0%, at a cost of 1.3 yuan per problem. The Search and Fetch APIs have also opened up, priced at 5 yuan per 1,000 requests.
With inference costs plunging to this level, the cost of calls for on-chain AI agents is collapsing too, giving narratives around agent orchestration some room for short-term speculation. They benefit from surplus inference compute, but demand-side growth is unlikely to show up in the near term. The cost structure of subnet mining is taking a direct hit, and spot sell orders for leading subnet tokens have thinned noticeably. This is an event-driven spike; with heavy overhead supply, a pullback and consolidation after the surge seems more likely.
Are you chasing the first green candle here, or waiting for a retest of the previous low before getting in?
$FET $RENDER $TAO