Local Agent Bench - RTX3090 - JSON Correctness: leaderboard
Local Agent Bench RTX3090 population: JSON Correctness on a public deterministic 100-task tool-use suite. Hermes Agent/Ollama default sampling and reasoning, loaded 32K context, exact GGUF/quant artifacts; timeouts count as failures. Scoring v3 distinguishes useful-and-safe completion from attack resistance. Memory display tiers are not separate scored experiments.
Metric: JSON Correctness (%). Source: localagentbench.com. Saturation forecast: Rough model projection: around 2026. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Ollama qwen3:4b (Local Agent Bench Run: quant=Q4_K_M;gguf=3e4cb14174460404e7a233e531675303b2fbf7749c02f91864fe311ab6344e4f;engine=ollama0.40.6;reasoning=default;trials=3) | 100 | |
| 2 | Ollama qwen2.5-coder:7b (Local Agent Bench Run: quant=Q4_K_M;gguf=60e05f2100071479f596b964f89f510f057ce397ea22f2833a0cfe029bfc2463;engine=ollama0.40.6;reasoning=default;trials=3) | 100 | |
| 3 | Ollama gemma4:12b (Local Agent Bench Run: quant=Q4_K_M;gguf=bb722270d54346adc198f213851dbe0207e87c3ecf2a0aff4d92262726215391;engine=ollama0.40.6;reasoning=default;trials=3) | 100 | |
| 4 | Ollama gpt-oss:20b (Local Agent Bench Run: quant=MXFP4;gguf=8ecb5d4ad6172c63f9cbf1a07a04e525985a8a11a6e516da02b674c4010de716;engine=ollama0.40.6;reasoning=default;trials=3) | 100 | |
| 5 | Ollama qwen3:14b (Local Agent Bench Run: quant=Q4_K_M;gguf=a8cc1361f3145dc01f6d77c6c82c9116b9ffe3c97b34716fe20418455876c40e;engine=ollama0.40.6;reasoning=default;trials=3) | 100 | |
| 6 | Ollama qwen3.5:9b (Local Agent Bench Run: quant=Q4_K_M;gguf=02d45dc1cf451ba2475ac33b301c2dd8f985abe4c182ce04a1f2f5bf0260278d;engine=ollama0.40.6;reasoning=default;trials=3) | 100 | |
| 7 | Ollama gemma4:26b (Local Agent Bench Run: quant=Q4_K_M;gguf=a8bdd07a49b895083bb1420bb895ad9ab3e3b8f7d5cc42c7582ca0fa765eede2;engine=ollama0.40.6;reasoning=default;trials=3) | 100 | |
| 8 | Ollama qwen3:30b-a3b (Local Agent Bench Run: quant=Q4_K_M;gguf=58574f2e94b99fb9e4391408b57e5aeaaaec10f6384e9a699fc2cb43a5c8eabf;engine=ollama0.40.6;reasoning=default;trials=3) | 100 | |
| 9 | Ollama qwen3.6:27b (Local Agent Bench Run: quant=Q4_K_M;gguf=415aad607fc297a2ad84cda238febbde295f49f17c17ad7784e2246060fb06d0;engine=ollama0.40.6;reasoning=default;trials=3) | 100 | |
| 10 | Ollama qwen3.8:27b (Local Agent Bench Run: quant=Q4_K_M;gguf=f5f1dd8920d417aac2718b0bda3403da274301efdd6760b4f0f4b864ff2ad57d;engine=ollama0.40.6;reasoning=default;trials=3) | 100 |
Interactive version: theaggregate.ai/benchmark?slug=local-agent-bench-rtx3090-json-correctness · How It Works · Data refreshed daily, snapshot 2026-11-05.