Mercury 2.5 leads small-model speed ranking

Mercury 2.5 tops modelgrep’s October 2026 ranking of compact AI models, delivering 830 tokens/sec at $0.040 per million input tokens. The leaderboard ranks 25 models by output speed, focusing on efficient models that trade some raw intelligence for lower costs and higher throughput.

Nemotron 3.5 Super VL (free) follows at 457 t/s, while Gemini 3.5 Flash Lite takes the next position at 409 t/s. Another free option, Nemotron 3.5 Lightning (free), delivers 300 t/s. The lineup spans open and proprietary models, with entries offering capabilities such as reasoning, tool use, structured JSON output and vision.

The ranking targets workloads including chat, classification and extraction, where speed and cost matter at high volume. Its ordering reflects output throughput rather than overall intelligence. Pricing, intelligence scores and time to first token are also listed, providing separate measures for comparing models beyond generation speed.