Inkling 2 Mini releases open weights, using a recurrent architecture for local inference
What happened
On November 21, Hacker News' front page reported the release of open weights for Inkling 2 Mini. It uses a 26.4B-parameter MoE with a recurrent Transformer architecture, takes about 20GB of memory, and runs its middle layers three times per token. The publisher reports 48.2% on DeepSWE v1.1 and 46.5% on Vals Index v2.2, and says multi-turn dialogue and multi-constraint instructions trail Qwen3.8-27B. Later that day, The Decoder reported the publisher is Thinking Machines, and that the open model solves problems with fewer tokens by running some Transformer layers recurrently and runs on a laptop.
Written by AI from the coverage · updated yesterday
Coverage
Follow the reports to see the story from different sides.
- The DecoderNew open model Inkling 2 Mini thinks in loops to solve problems with fewer words and still runs on a laptop
Thinking Machines 发布 Inkling 2 Mini,这款开放 MoE 模型通过循环运行部分 Transformer 层,以更少 token 求解问题。
- Hacker News front pagePickInkling 2 Mini releases open weights, using a recurrent architecture for local inference
Inkling 2 Mini is out with open weights: a 26.4B-parameter MoE built on a recurrent Transformer, at roughly 20GB of total memory. Each token runs the middle layers three times. The publisher reports 48.2% on DeepSWE v1.1 and 46.5% on Vals Index v2.2, and says the model trails Qwen3.8-27B on multi-turn dialogue and multi-constraint instructions.
Heat over time
Heat now 7·Comparable peak 16(Nov 21 19:00)·Comparable change over 24 hours -50%
The trend only compares accounts observed without gaps, so its range may be smaller than the current heat. Hover or tap the chart for each hour; the left and right arrow keys step through it.