Terminal Bench 2.1
A benchmark for evaluating the ability of large models to execute tasks in terminal and command-line environments.
Timeline (last 90 days)
Leaked data shows Gemini 4 Pro scoring 88.7% on DeepSwe V1.1, 95.3% on Terminal Bench 2.1, 86.8% on OS World 2.0, and 20% on GDeepVal AAVR, with context expanded to 2 million tokens and pricing at $2.25 per million tokens for input and $11.25 for output.
The test used Terminal Bench 2.1 as the terminal interaction task set, running two model lines: top models connected via Claude Code and DeepSeek connected via Open Code through OpenRouter.
Under the DeepSuite V1.1 benchmark, the same model's scores fluctuate from 69.8% to 74.2% across different harnesses.
According to a content creator's hands-on testing, GLM-5.3 Flash scored close to Claude Opus 4.8 on the Terminal Bench 2.1 complex terminal task test.
DeepSeek Harness Minimal + V4 Pro achieves 87.9% on Terminal Bench 2.1.
All Facts (5)
Leaked data shows Gemini 4 Pro scoring 88.7% on DeepSwe V1.1, 95.3% on Terminal Bench 2.1, 86.8% on OS World 2.0, and 20% on GDeepVal AAVR, with context expanded to 2 million tokens and pricing at $2.25 per million tokens for input and $11.25 for output.
50%UnverifiedThe test used Terminal Bench 2.1 as the terminal interaction task set, running two model lines: top models connected via Claude Code and DeepSeek connected via Open Code through OpenRouter.
50%UnverifiedUnder the DeepSuite V1.1 benchmark, the same model's scores fluctuate from 69.8% to 74.2% across different harnesses.
50%UnverifiedAccording to a content creator's hands-on testing, GLM-5.3 Flash scored close to Claude Opus 4.8 on the Terminal Bench 2.1 complex terminal task test.
50%UnverifiedDeepSeek Harness Minimal + V4 Pro achieves 87.9% on Terminal Bench 2.1.
50%