GPT-5 (high) on Terminal-Bench (Terminus-2): 42.5% accuracy

The LayerLens Team covers AI model evaluations, benchmark analysis, and the evolving landscape of AI performance. For the latest independent evaluation data, explore Stratix.

Summary

GPT-5 (high) from OpenAI scored 42.5 on Terminal-Bench (Terminus-2), placing it top 10 (rank 10 of 21) on this benchmark. This places the model in the below-frontier band for Terminal-Bench (Terminus-2). Acceptable for cost-sensitive workloads or as part of a multi-model ensemble. Not a default choice for high-stakes routing.

Model details

• Provider: OpenAI

• Model key: openai/gpt-5-high

• Context length: 400,000 tokens

• License: Proprietary

• Open weights: no

Benchmark methodology

Benchmark goal: Evaluate the LLM's ability to solve a diverse set of programming and reasoning tasks.

Scoring metrics:

• Score: Binary success (1) or failure (0) based on task completion and correctness.

• Duration: Time taken to complete the task (in seconds).

Analysis

Key takeaways:

• The LLM demonstrates proficiency in completing individual programming and environment setup tasks, but struggles with complex, multi-stage problems.

• Planning and Robustness in adapting to unseen dependencies remain significant challenges.

• The model shows inconsistent performance in code generation tasks, failing to develop robust algorithms or adapt to environmental constraints.

Failure modes observed

Common failure modes:

• Failure to follow complex instructions involving multiple tools and configurations.

• Incomplete or incorrect code generation leading to non-functional solutions.

• Not able to plan multistep task to arrive at expected goal (e.g. Creating a website profile from info.md)

• Not able to reengineer binaries with exact same behavior

• Dependency errors: Not able to fix fasttext installation

Example: Failing to build a complete end-to-end system, such as setting up a Git server with deployment or creating a fully functional web scraper. In the Linux Kernel building, the agent fails to follow the custom configuration.

Secondary metrics

• Readability score: 0.0

• Toxicity score: 0.000

• Ethics score: 0.000

Run this evaluation yourself

Stratix evaluates GPT-5 (high) continuously across 11+ benchmarks. To replicate this Terminal-Bench (Terminus-2) evaluation on your own model, traces, or a different benchmark configuration, open the model in Stratix.

Source: Stratix evaluation 68fff48651dd2ed6199046b2. Updated 2025-10-28.