ISA-Bench - Human Resource Machine: leaderboard

Metric: Tasks solved (%; share of the 36 Human Resource Machine tasks (an accumulator machine with a small instruction set) whose program passes every verification case in the game's own parser, virtual machine and verifier; up to five attempts with parse, runtime or per-test execution feedback; Ollama default builds, temperature 0.7, 4,096 output tokens). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 14 models tracked.

Top models

#ModelScore
1ISA-Bench Gemma-4 31B (Thinking, Ollama build)88.89
2ISA-Bench Phi-4-Reasoning (Ollama build)55.56
3ISA-Bench Qwen3 32B (Thinking, Ollama build)44.44
4ISA-Bench Qwen3-Coder 30B (Ollama build)38.89
5ISA-Bench Qwen3 14B (Thinking, Ollama build)36.11
6ISA-Bench DeepSeek-R1 32B (Ollama build)36.11
7ISA-Bench Devstral-Small-2 (Ollama build)33.33
8ISA-Bench Qwen3 8B (Thinking, Ollama build)33.33
9ISA-Bench Qwen2.5-Coder 32B (Ollama build)19.44
10ISA-Bench Granite-4.1 8B (Ollama build)16.67

Interactive version: theaggregate.ai/benchmark?slug=isa-bench-human-resource-machine · How It Works · Data refreshed daily, snapshot 2026-09-26.