New Cognition model took Inkling-Small, a 276B open-weight model with only 12B active…
This is a dev post classified by AGK Intelligence as AI dev tools (a launch), kept by the Dev Radar because it carries real work, not commentary.
New Cognition model took Inkling-Small, a 276B open-weight model with only 12B active parameters, and turned it into a coding beast. 👀 ... but the improved weights aren't public. Maybe we can just ask Fable to reverse-engineer it? 😃 This is Cognition's SWE-2 Flash. Stats 🧠 276B total / 12B active ⚡ 3× faster than Cognition's giant SWE-2 💰 5× cheaper to serve, according to Cognition Same model size, but here are the improvements (kind of crazy) DeepSWE 1.1: Inkling-Small → 17.4 🔥 SWE-2 Flash → 60.1 FrontierCode 2.0: 3.2 → 24.0 Terminal-Bench 2.1: 64.7 → 89.2 SWE-Bench Pro v2 Hard: 56
Posted by David Hendrickson (12.7k followers) 1 h ago · 0 likes · 186 views · view the original post on X. Kept by the Dev Radar as AI dev tools.
More dev work like this
- Grok Build v1.0.76 is out with a few practical updates. Standouts include ctrl+L redraws… — @blankspeaker
- Clover-35B-A3B, a diffusion LLM, decodes at 25.8 tok/s in BF16 on one DGX Spark (128 GB) — @niklaslenz_ai
- Give Claude Code a Facebook Page workflow, not another blank prompt. — @DanKornas
- How can larger coding models, longer context windows, and faster inference fit on one… — @windowsdev
- Meet Qwen3.5-0.8B weight-space programming: a tiny but mighty image-text-to-text model.… — @HuggingModels
- 𝕏 just rebuilt its entire web app from scratch, and the performance improvements are… — @XFreeze
- One of the use cases I found for @DevinAI is that it can help you if you plan to train… — @Chris_Wozniczek
- OpenDataLoader PDF 把 PDF 转成 Markdown 的速度，能做到一页 0.015 秒，纯 CPU 跑得动，还开源免费。 — @bkdgiffug
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 31.5k posts from 5k X accounts over the last 21 days, 9.3k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-04 23:13 UTC. Full method.