VQV Signal

RESEARCH WATCH TECHNICAL

Inkling 2 Mini and Qwen 3.8 27B tie at 20.1 on AA index, per X post

According to a post on X by @sudoingX, the model card for Thinking Machines' Inkling 2 Mini shows it tied with Qwen 3.8 27B at 20.1 on the AA index. The post says Inkling is ahead on some coding benchmarks but behind on GDPval-AA and non-hallucination, and that a Ghidra reverse-engineering test had it trailing Qwen. A separate X post by Thinking Machines' @cHHillee says most of the model's training took under 10% of one researcher's time, plus agents. The post also mentions improving local runtimes, though the snippet cuts off before details.

Source: Scobleizer News Pulse (X) · x.com Found via Robert Scoble’s Scobleizer News Pulse · a post on X, not independently verified Published 2026-11-21T06:25:00+00:00 Detected 2026-11-21T16:05:28+00:00
View original source

According to a post on X by @sudoingX, the model card for Thinking Machines' Inkling 2 Mini shows it tied with Qwen 3.8 27B at 20.1 on the AA index. The post says Inkling is ahead on some coding benchmarks but behind on GDPval-AA and non-hallucination, and that a Ghidra reverse-engineering test had it trailing Qwen. A separate X post by Thinking Machines' @cHHillee says most of the model's training took under 10% of one researcher's time, plus agents. The post also mentions improving local runtimes, though the snippet cuts off before details.

AI-assisted summary based on the listed source.

@sudoingX notes the model card shows the two tied at 20.1 on the AA index, with Inkling ahead on some coding benchmarks but behind on GDPval-AA and non-hallucination, and a Ghidra reverse-engineering test had it trailing Qwen; Thinking Machines' @cHHillee says most of its training took under 10% of one...

Developers choosing open models for coding or local use may find the gap is small rather than a clear lead, so results on their own tasks matter more than headline scores. The reported training effort is also worth tracking for teams weighing how cheaply capable models can be built. Watch for fuller benchmark tables and further independent evaluations before drawing firm conclusions.

Signal Strength 95% Technical label WATCH Public Interest 54 Category RESEARCH Reader Depth TECHNICAL

Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.

Public Interest components
Recognizable Entity Score 90 Practical Impact Score 8 Novelty Interest Score 94 Consequence Score 34 Curiosity Score 16 Shareability Score 46

VQV surfaced this signal because it is recent, relevant to Open Source LLMs, connected to Scobleizer News Pulse (X).