Rohan Paul@rohanpaul_ai
57AI 编辑部评分,满分 100
2026-07-31 09:52· 1小时前
跳到正文
AI 摘要

Thinking Machines 发布开源权重模型 Inkling-Small,总参数量 276B、每 token 激活 12B,仅为 Inkling 激活参数(41B)的三分之一。该模型在 HLE 上以 31.6% 超过更大的 Inkling(29.7%),SWE-bench Verified 达 80.2%,并针对音频智能优化了预训练混合、知识蒸馏及两周智能体编码 RL。

Thinking Machines released open-weight model Inkling-Small, cuts Inkling's active parameter count from 41B to 12B per token.

Pushed ahead of the larger Inkling on HLE, 31.6% versus 29.7%, and to 80.2% on SWEBench Verified, while activating less than one-third as many parameters.

The model also received a revised pretraining mix, distillation from Inkling, and two additional weeks of agentic coding RL.

They said that they crafted Inkling-Small for audio intelligence, making it a good candidate for real-world audio applications.

Thinking MachinesToday, we are releasing Inkling-Small. Inkling-Small achieves comparable performance to Inkling at a quarter of its size. It features 276B total parameters, 12B...
Rohan Paul · @rohanpaul_ai · X·2026-07-31 09:52·1小时前
在 X 看原推· x.com
AI 摘要

Thinking Machines 发布开源权重模型 Inkling-Small,总参数量 276B、每 token 激活 12B,仅为 Inkling 激活参数(41B)的三分之一。该模型在 HLE 上以 31.6% 超过更大的 Inkling(29.7%),SWE-bench Verified 达 80.2%,并针对音频智能优化了预训练混合、知识蒸馏及两周智能体编码 RL。

Thinking Machines released open-weight model Inkling-Small, cuts Inkling's active parameter count from 41B to 12B per token.

Pushed ahead of the larger Inkling on HLE, 31.6% versus 29.7%, and to 80.2% on SWEBench Verified, while activating less than one-third as many parameters.

The model also received a revised pretraining mix, distillation from Inkling, and two additional weeks of agentic coding RL.

They said that they crafted Inkling-Small for audio intelligence, making it a good candidate for real-world audio applications.

Thinking MachinesToday, we are releasing Inkling-Small. Inkling-Small achieves comparable performance to Inkling at a quarter of its size. It features 276B total parameters, 12B...
在 X 查看原推x.com