IBM Granite 4 Nano's browser-local deployment beats Qwen3-1.7B on IFEval — sub-2B parameter edge-tier procurement default emerges around Granite + Mistral Small 4
IBM's Granite 4 Nano models (small enough to run locally in-browser) score 78.5 on IFEval, beating Qwen3-1.7B (73.1) and other 1-2B competitors. Combined with Mistral Small 4's 6B-active Apache-2.0 release, the sub-2B + sub-7B edge-deployment tier now has clear category leaders for browser-local and on-device inference procurement.
The substantive piece is the edge-tier procurement consolidation. Sub-2B-parameter models occupy a distinct procurement category from frontier OSS — they target browser-local inference, on-device deployment, and resource-constrained edge scenarios where running a 30B model is infeasible. Granite 4 Nano beating Qwen3-1.7B on instruction-following at this parameter scale matters because the edge-tier capability has been the weakest segment of the OSS market; Granite 4 Nano demonstrates that sub-2B models can deliver near-frontier instruction-following quality.
The structural read for the OSS-frontier procurement segmentation is that two-tier OSS deployment becomes the H2 2026 default: heavy reasoning workloads run on DeepSeek V4-Pro / MiniMax M3, edge/latency-sensitive workloads run on Granite 4 Nano / Mistral Small 4. The middle tier (7B-13B dense) hollows out as buyers pick from the top or the bottom of the OSS stack rather than the middle.
VentureBeat — IBM's open source Granite 4.0 Nano AI models are small enough to run locally directly in your browser → · Featherless — Best Open-Source LLMs in 2026 → · Kilo.ai — Best Open-Source & Open-Weight Coding Models (2026) →