Gguirgsilvainguilsilva.dev·Apr 11 · 5 min readCRIT Framework and Model Routing: Spending Less on AI Without Losing QualityIn the previous articles I covered the SWE-Bench numbers, the open-weight models worth knowing, and why bigger context doesn't mean better output. Now I want to connect everything with two things I us10
Gguirgsilvainguilsilva.dev·Apr 9 · 5 min readA Bigger Context Window Doesn't Mean Better QualityWhen a company says their model has 1 million tokens of context, everyone assumes bigger is better. I thought the same thing. But after running LLMs in production for a while, I found out it's not tha00
Gguirgsilvainguilsilva.dev·Apr 6 · 4 min readMiniMax M2.5 and Qwen3.5: The Open-Weight Models Worth Knowing AboutIn the last article I showed the SWE-Bench numbers. Open-weight models are basically tied with the proprietary ones now. Two models stood out to me: MiniMax M2.5 and Qwen3.5. Here's what I found out a00
Gguirgsilvainguilsilva.dev·Apr 4 · 3 min readOpen-Weight vs Proprietary: What SWE-Bench Verified Is Telling Us in 2026I was looking at the SWE-Bench Verified leaderboard last week and the numbers surprised me. The gap between proprietary and open-weight models is almost gone. Not in some academic test. In actual bug 00
