Inkling 2 Mini Beats Qwen 27B on a Single RTX 4060, Runs 3.5x Faster
LiTianleli · x · 2026-11-22
User analogalok ran Inkling 2 Mini locally on a single RTX 4060 (8GB VRAM) with 16GB DDR4 RAM, claiming it beats Qwen 4 27B on DeepSWE 1.1.
- Model: Unsloth Inkling-2-Mini-UD-Q3KXL.gguf, medium reasoning budget, 24k context with Q8 KV cache
- Test: one-shot prompt to build a detailed realistic car in a single HTML file with Three.js, CDN allowed
- Result: roughly 3.5x faster than Qwen 3.8 27B Dense (Q4KXS) on the same rig, with solid initial output before raising the reasoning budget
Full benchmarks, llama.cpp flags and comparisons are promised soon.
More from Infra
- SGLang runs Inkling 2 Mini in 65GB on DGX Spark, 2x faster prefill than vLLM — AIFlow_ML · 2026-11-22
- Open-source Agent Skill hill-climbs local LLM serving speed, users report up to 5x gains — CatAstro_Piyush · 2026-11-22
- "MoE is a datacenter architecture; looped transformers are a laptop architecture" — andreisavu · 2026-11-22
- H100 rental prices peaked in June; token prices keep falling while compute stays robust — AIFlow_ML · 2026-11-22
- LiquidAI's LFM3 runs on your phone; Nikita Bier's local-model thesis resurfaces — maximelabonne · 2026-11-22
- Chutes drama: TEE-encrypted inference flagged on OpenRouter despite being 'most secure' — const_reborn · 2026-11-22