Pokee-Isaac 28B Report - Terminal-Bench 2.1 (86 text tasks): leaderboard
Terminal-Bench 2.1 containerized command-line tasks in software engineering, machine learning, security and system administration, with Harbor 0.20.0 and the Terminus-2 agent for every model: the 86 text-only tasks of 89, one attempt each, a 100-turn cap and 16x default timeouts; errored trials fail. Tasks resolved (%). From Pokee AI's Pokee-Isaac 28B technical report (September 10, 2026 revision): Pokee ran every listed model itself through hosted APIs under one harness, with each peer's default temperature (unless a benchmark section fixes one) and a high thinking level where available; exhausted retries, context overflows and verifier failures count as zero. Higher is better.
Metric: Tasks resolved (%). Source: pokee.ai. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Luna | 69.8 | #117 |
| 2 | Qwen 3.5 122B A10B | 46.5 | #329 |
| 3 | Gemini 3.5 Flash Lite | 46.5 | #300 |
| 4 | Claude Haiku 4.5 | 34.9 | #568 |
| 5 | Nemotron 3 Super | 24.4 | #766 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=pokee-isaac-28b-report-terminal-bench-2-1-86-text-tasks · How It Works · Data refreshed daily, snapshot 2026-11-12.