Terminal-Bench 3.0 leaderboard
As of November 8, 2026, DeepSeek V4.1 Flash (DeepSeek) leads Terminal-Bench 3.0 at 30%, ahead of GLM-5.3 at 28.3% and Grok 4.6 at 26%. 9 systems ranked. Every score here is published by the model's own lab; each row links to the page it comes from.
What is Terminal-Bench 3.0? Terminal-Bench 3.0 is a harder, longer-running set of terminal tasks than version 2, covering databases, machine learning, hardware design, formal proofs, CAD, science and business processes. Made by The Terminal-Bench and Harbor team (Laude Institute), July 2026.
Top models on Terminal-Bench 3.0, November 2026
| # | Model | Score | Price / 1M | Context | Source |
|---|---|---|---|---|---|
| 1 | DeepSeek V4.1 Flash DeepSeek | pass@1 | $0.3 / $1.2 | 1M | Vendor-reported DeepSeek |
| 2 | GLM-5.3 Zhipu (Z.ai / GLM) | as reported | $1.4 / $4.4 | 1M | Vendor-reported Z.ai (Zhipu) |
| 3 | Grok 4.6 SpaceXSI | high effort | $2 / $6 | 500K | Vendor-reported SpaceXSI (xAI) |
| 4 | Grok 4.5 SpaceXSI | high effort | $2 / $6 | 500K | Vendor-reported SpaceXSI (xAI) |
| 5 | Gemini 3.7 Flash Deprecated Google DeepMind | as reported | $0.75 / $3.75 | 1M | Vendor-reported Google DeepMind |
| 6 | DeepSeek V4 Pro DeepSeek | pass@1 | $1.32 / $3.96 | 1M | Vendor-reported DeepSeek |
| 7 | DeepSeek V4 Flash Deprecated DeepSeek | pass@1 | $0.44 / $1.32 | 1M | Vendor-reported DeepSeek |
| 8 | Gemini 3.6 Flash Google | as reported | $0.75 / $3.75 | 1M | Vendor-reported Google DeepMind |
| 9 | GLM-5.2 Zhipu (Z.ai / GLM) | as reported | $1.4 / $4.4 | 1M | Vendor-reported Z.ai (Zhipu) |
Terminal-Bench 3.0: score against price
Can you trust Terminal-Bench 3.0 scores?
No independent leaderboard publishes Terminal-Bench 3.0 results for these models yet, so every number on this page is the lab's own report of its own model. Labs choose the harness, the effort level and the subset they report, and they rarely publish failures. Read the setting under each score before comparing two rows.
- Agents have open internet access and are only told not to look up task solutions, so leakage depends on compliance.
- Version 4.0 replaced 3.0 about a month later with fixed and removed tasks; 3.0 scores are tied to that version.
- Results are per agent and model pair: the harness is part of the score.
How Terminal-Bench 3.0 works
The task
The agent works in a container (sometimes several, sometimes with a GPU) and has to produce an artifact; the artifact is copied to a separate verifier container for checking.
The score
Each trial is a pass, a fail or an error, and the headline is the pass rate.
Test set
74 tasks, as published by the authors.
Made by
The Terminal-Bench and Harbor team (Laude Institute), first published July 2026.
Official sources
- www.tbench.ai/news/terminal-bench-3-0
- Definition read on 2026-11-08
Sources and freshness
9 published scores from 4 sources. Each was read on its source page; the oldest reading on this page is from November 8, 2026. Prices beside each score come from our model price index and move when a vendor reprices.
- Vendor-reported DeepSeek
- Vendor-reported Google DeepMind
- Vendor-reported SpaceXSI (xAI)
- Vendor-reported Z.ai (Zhipu)
Terminal-Bench 3.0 FAQ
Answers derived from the 9 published scores on this page and the current price index.