Inkling-2-mini (~25B) runs on one RTX 3090, early test says it beats bigger MoE model
LiTianleli · x · 2026-11-22
A user tested thinkymachines' Inkling-2-mini: a 25B model running locally on a single RTX 3090 with the Unsloth UD-Q4KXL quant. Asked to one-shot build a floating island in snowy weather (no harness, no tool calls), it used only 23,600 tokens (2,891 thinking + 20,709 answer) for a high-quality result. The author compared it against Hy4-Flash via API and found Mini's output clearly better — notable news for local AI, though more testing is needed.
More from Models
- Tesla starts rolling out FSD v15 with 25x better safety-critical handling — ArnoCandel · 2026-11-22
- Joke tweet suggests Grok 4.5 vs 4.7 teams 'need to be studied' — zdhnarsil · 2026-11-22
- Simon Willison: Gemini 4 Hydrogen 'xhigh' thinking costs 6x more but is far better — film_girl · 2026-11-22
- Researcher argues LLM companies have no moat as Gemini 4 Neon kills the data flywheel thesis — adam_karvonen · 2026-11-22
- Anthropic releases FlashProve with Lean formalization, proving 400B MoE inference on one H100 — jiqizhixin · 2026-11-22
- Gemini users say Low thinking still burns 5-hour quotas on basic prompts, want a non-thinking option back — patg84 · 2026-11-22