Thinking Machines 发布开放权重模型 Inkling 2 MiniChinese only
Inkling 2 Mini 可处理文本和图像输入,提供 100 万 token 上下文窗口,并以 Apache 2.0 许可发布。Artificial Analysis 的评测显示,它在 Intelligence Index 上与 Qwen3.8 27B(xhigh)得分相当;按 Thinking Machines 的定价,每项评测任务花费 0.59 美元、耗时 9.0 分钟,低于 Qwen3.8 27B 的 3.53 美元和 38.2 分钟。
1 developments2 reports2 sources3 verified facts
Verified facts 3
Thinking Machines Inkling 2 Small: the highest scoring open weights model outside of China pending weights release November 5, 2026
OpenAI发布了GPT-6.2 Astra和GPT-6.2 Sol,报道称其中一款模型在Artificial Analysis Intelligence Index中排名第二。
Event developments 1
Thinking Machines releases Inkling 2 Mini with open weights
Inkling 2 Mini accepts text and image input, offers a 1M-token context window, and is released under an Apache 2.0 license. In Artificial Analysis's evaluation, it matches Qwen3.8 27B (xhigh) on the Intelligence Index; at Thinking Machines' pricing, each evaluation task costs $0.59 and takes 9.0 minutes, compared with $3.53 and 38.2 minutes for Qwen3.8 27B.
View source reports · 2 sources · 2 reports
- Artificial Analysis 模型评测 · Thinking Machines releases Inkling 2 Mini with open weights
- ginobefun (@hongming731) · Thinking Machines releases open-weight Inkling 2 Mini with a laptop demo
All reports 2
Thinking Machines releases open-weight Inkling 2 Mini with a laptop demo
Inkling 2 Mini combines repeated execution of intermediate layers with native low-precision training. Thinking Machines also published coding evaluations and a laptop demo. For local development, the brief suggests assessing model size, computation per token, and whether producing less output reduces task completion time.
1 more source reported this
- Artificial Analysis 模型评测
Thinking Machines releases Inkling 2 Mini with open weights
Inkling 2 Mini accepts text and image input, offers a 1M-token context window, and is released under an Apache 2.0 license. In Artificial Analysis's evaluation, it matches Qwen3.8 27B (xhigh) on the Intelligence Index; at Thinking Machines' pricing, each evaluation task costs $0.59 and takes 9.0 minutes, compared with $3.53 and 38.2 minutes for Qwen3.8 27B.
1 more source reported this
- ginobefun (@hongming731)