OBVIOUSBENCH · PUBLIC MODEL RELIABILITY
A small, evidence-first library drawn from the 432 public configurations in the 2026-07-25 release. It complements—rather than replaces—the complete interactive field.
FASTEST PATH TO A RELIABILITY BAR
Each table picks the cheapest observed public configuration from each model family that clears the stated answer pass3 threshold, then ranks those family configurations by cost. Open any row in the complete global explorer for the full field and Pareto context.
Lowest-cost observed configuration per model family clearing this threshold.
| Model family | Setting | Pass3 | Cost |
|---|---|---|---|
| Gemma 4 31B IT | High | 100% | $0.0251 |
| Hy3 | Low | 99.3% | $0.05838 |
| Gemini 3 Flash Preview | Medium | 99.3% | $0.2473 |
| qwen3.5 27B | Reasoning | 100% | $0.2728 |
| qwen3.5 397B A17B | Reasoning | 99.3% | $0.4429 |
| Gemini 3.5 Flash | Low | 99.3% | $0.5007 |
| Claude Opus 5 | Medium | 99.3% | $0.503 |
| o3 | Medium | 100% | $0.5356 |
Lowest-cost observed configuration per model family clearing this threshold.
| Model family | Setting | Pass3 | Cost |
|---|---|---|---|
| Gemma 4 31B IT | High | 100% | $0.0251 |
| Gemma 4 26B A4B IT | Medium | 96.5% | $0.03507 |
| Ring-2.6-1T | Default | 97.9% | $0.05153 |
| GPT 5 nano | Medium | 95.1% | $0.05523 |
| Hy3 | Low | 99.3% | $0.05838 |
| Step 3.5 Flash | Default | 95.1% | $0.06471 |
| Seed-2.0-Mini | Default | 95.1% | $0.08202 |
| Gemini 3.1 Flash Lite | Medium | 95.8% | $0.1175 |
Lowest-cost observed configuration per model family clearing this threshold.
| Model family | Setting | Pass3 | Cost |
|---|---|---|---|
| GPT OSS 120B | Medium | 94.4% | $0.008422 |
| GPT OSS 20B | High | 92.4% | $0.01742 |
| Laguna XS 2.1 | Default | 92.4% | $0.02004 |
| GPT 5 nano | Low | 91% | $0.02441 |
| Gemma 4 31B IT | High | 100% | $0.0251 |
| qwen3 235B A22B 2507 | Reasoning | 94.4% | $0.03026 |
| Gemma 4 26B A4B IT | Medium | 96.5% | $0.03507 |
| qwen3 14B | Reasoning | 91.7% | $0.04244 |
START WITH A LAB
These eight shortcuts are editorial navigation, not an exclusion of the other manufacturers in the public field.
CURATED DECISION PAGES
Each page shows the observed settings, exact pass3 and cost values, the release source, and a share-ready image. It is a purposeful sample, not a bulk index of every model name.
COMPLETE FIELD
Every public manufacturer is listed below. The homepage data explorer remains the authoritative way to inspect all 432 configurations and change filters.
