OK, aside from Thibaut's game, there's other news, I tell you. Mira Morti's think tank has released the Inkling 2 Small: the model is already available on Tinker to promote further training, and the company plans to place weights for Apache 2.0 next week, after external security testing. Within MoE, 17 billion active out of 594, there is image input and personalized effort. According to the company table, the model is inferior to DeepSeek V4.1 Flash, but surpassed GLM-5.3-Flash: 72.1% on DeepSWE v1.1 against 77.0% and 63.4%. The company also confirmed that the Fledge-Alpha Stealth model in OpenCode was an early barrier of 2 small barriers. Among the weaknesses, the authors name ectopics, mistakes in choosing tools in repositories with a large number of skills and a weak response in chat. Channel @neuro.

OK, aside from Thibaut's game, there's other news, I tell you. Mira Morti's think tank has released the Inkling 2 Small: the model is already available on Tinker to promote further training, and the company plans to place weights for Apache 2.0 next week, after external security testing. Within MoE, 17 billion active out of 594, there is image input and personalized effort. According to the company table, the model is inferior to DeepSeek V4.1 Flash, but surpassed GLM-5.3-Flash: 72.1% on DeepSWE v1.1 against 77.0% and 63.4%. The company also confirmed that the Fledge-Alpha Stealth model in OpenCode was an early barrier of 2 small barriers. Among the weaknesses, the authors name ectopics, mistakes in choosing tools in repositories with a large number of skills and a weak response in chat. Channel @neuro.