Agents' Last Exam
An agent benchmark reported in DeepSeek's V4 Flash 0731 launch comparison.
Top models on Agents' Last Exam — November 12, 2026
As of November 12, 2026, GPT-6.1 Terra leads the Agents' Last Exam leaderboard with 60.0% , followed by GPT-6 Astra (59.3%) and GPT-6 Sol (56.4%).
GPT-6.1 Terra
OpenAI
Reported by OpenAIGPT-6 Astra
OpenAI
Reported by OpenAIGPT-6 Sol
OpenAI
Reported by OpenAI21 modelsAgentic3% of Agentic reference weightCurrentUpdated November 12, 2026
Leaderboard (21 models)
All 21 results on this page were reported by the labs that built the models.
| Rank | Model / configuration | Score | Who ran it | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | GPT-6.1 TerraOpenAIReported by OpenAI | 60.0% | Reported by OpenAI | Not reported | Closed |
| 2 | GPT-6 AstraOpenAIReported by OpenAI | 59.3% | Reported by OpenAI | Not reported | Closed |
| 3 | GPT-6 SolOpenAIReported by OpenAI | 56.4% | Reported by OpenAI | Not reported | Closed |
| 4 | Qwen3.8 MaxAlibabaReported by Alibaba | 52.4% | Reported by Alibaba | Not reported | Open |
| 5 | Qwen3.8-Flash-NextAlibabaReported by Alibaba | 51.2% | Reported by Alibaba | Not reported | Open |
| 6 | Kimi K3.1 ProMoonshot AIReported by Moonshot AI | 43.2% | Reported by Moonshot AI | Not reported | Open |
| 7 | Qwen3.8-27BAlibabaReported by Alibaba | 42.9% | Reported by Alibaba | Not reported | Open |
| 8 | Hy4TencentReported by Tencent | 41.0% | Reported by Tencent | Not reported | Pending |
| 9 | Gemini 4 ArgonGoogleReported by Google | 39.5% | Reported by Google | Not reported | Closed |
| 10 | LongCat-2.5MeituanReported by Meituan | 38.5% | Reported by Meituan | Not reported | Open |
| 11 | DeepSeek V4.1 Flash 1017DeepSeekReported by DeepSeek | 35.4% | Reported by DeepSeek | Not reported | Open |
| 12 | MiMo-V2.6-ProXiaomiReported by Xiaomi | 31.6% | Reported by Xiaomi | Not reported | Open |
| 13 | MiniMax M3.1 FlashMiniMaxReported by MiniMax | 29.6% | Reported by MiniMax | Not reported | Closed |
| 14 | Step 5 PreviewStepFunReported by StepFun | 29.5% | Reported by StepFun | Not reported | Open |
| 15 | GLM-5.3Z.AIReported by Z.AI | 28.5% | Reported by Z.AI | Not reported | Open |
| 16 | MiMo-V2.6-FlashXiaomiReported by Xiaomi | 27.6% | Reported by Xiaomi | Not reported | Open |
| 17 | Gemini 3.7 FlashGoogleReported by Google | 26.3% | Reported by Google | Not reported | Closed |
| 18 | GLM-5.3-FlashZ.AIReported by Z.AI | 26.3% | Reported by Z.AI | Not reported | Open |
| 19 | DeepSeek V4 Pro 0813DeepSeekReported by DeepSeek | 25.7% | Reported by DeepSeek | Not reported | Open |
| 20 | DeepSeek V4 Flash 0731DeepSeekReported by DeepSeek | 25.2% | Reported by DeepSeek | Not reported | Open |
| 21 | Hy4 previewTencentReported by Tencent | 22.8% | Reported by Tencent | Not reported | Open |
- Lab-reported
- Benchmark-run
- Independent
Results reported by model developersUnranked
Labs ran these themselves, with a metric, split or setting the table above does not use. They take no rank here.
| Model | Reported value | What was measured | Against the table above | Who ran it |
|---|---|---|---|---|
| Gemini 4 CarbonGoogle | 77.4%Averaged rubric score | Agents' Last Exam v1.1122-task release; the stored field is the binary pass rate | Not comparable to the scores above | Reported by Google |
| Gemini 4 HydrogenGoogle | 71.6%Averaged rubric score | Agents' Last Exam v1.1122-task release, Antigravity CLI, five-hour window; the stored field is the binary pass rate | Not comparable to the scores above | Reported by Google |
| Gemini 4 ArgonGoogle | 69.0%Averaged rubric score | Agents' Last Exam v1.1122-task release; not the binary pass rate | Not comparable to the scores above | Reported by Google |
A lab’s own result is shown for reference. It never takes a rank on this board.
How we label sourcesAccording to BenchLM.ai, GPT-6.1 Terra leads the Agents' Last Exam benchmark with a score of 60.0%, followed by GPT-6 Astra (59.3%) and GPT-6 Sol (56.4%). There is significant spread across the leaderboard, making this benchmark effective at differentiating model capabilities.
21 models have been evaluated on Agents' Last Exam. The benchmark falls in the Agentic category. BenchAlign v5.8 gives Agents' Last Exam 3% of the Agentic reference weight, so it moves the Agentic leaderboard and the overall ranking. Reference weights are relative weights in the calibrated model, not fixed shares of a score.
About Agents' Last Exam
Year
2026
Tasks
Agent tasks
Format
Provider-reported task score
Difficulty
Advanced agentic work
BenchLM stores provider-run values, such as DeepSeek's max-effort launch result, under the exact published benchmark label.
Freshness and provenance
Version
Agents' Last Exam 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does Agents' Last Exam measure?
An agent benchmark reported in DeepSeek's V4 Flash 0731 launch comparison.
Which model scores highest on Agents' Last Exam?
GPT-6.1 Terra by OpenAI currently leads with a score of 60.0% on Agents' Last Exam.
How many models are evaluated on Agents' Last Exam?
21 AI models have published results on Agents' Last Exam in the BenchLM catalog.
Compare top models on Agents' Last Exam
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.