Nanbeige4.2-3B launches as a 3B Looped Transformer model that beats larger baselines
Wooden-Deer-1276 · reddit · 2026-07-22
Nanbeige4.2-3B debuts as a 3B agentic model with a Looped Transformer
- The model is built on Nanbeige4.2-3B-Base and uses a Looped Transformer architecture that reuses layers to increase capacity without adding parameters.
- With only 3B non-embedding parameters, it is positioned as a compact model for agentic behavior plus broad reasoning and alignment.
- The benchmark chart shows Nanbeige4.2-3B outperforming larger baselines such as Gemma4-E4B, Gemma4-12B, Qwen3.5-4B, and Qwen3.5-9B on a mix of agent and reasoning tasks, including MCP-atlas, PinchBench-v2, SWE-bench Pro, Terminal Bench 2.0, GPQA-Diamond, and HMMT-Feb-2026.
- The post claims solid performance on both general-agent and code-agent tasks despite the compact size.
Related event: Nanbeige4.2-3B Released with Looped Transformer for Agent Capabilities(5 posts)→
More from Models
- Kimi K3.1 Pro tops MathArena open models, closing frontier gap to about 6 months — teortaxesTex · 2026-10-26
- Ruff creator Charlie Marsh: frontier models find infinite bugs in any software — charliermarsh · 2026-10-26
- Dev testing one skill across Opus 5.5 and Sol 6.1 finds instruction following 'all over the map' — lucasmeijer · 2026-10-26
- supermemory unveils learner-1, claiming continual learning for any model — julianweisser · 2026-10-26
- Anthropic says Opus 5.5 cuts cost per task by ~40% vs Opus 5 — taufiqintech · 2026-10-26
- AI search APIs ranked: top 12 within 7 points, yet prices differ nearly 7x — FinanceYF5 · 2026-10-26