CorpBench Work
Agents carry out 100 business workflows across finance, marketing, operations, RevOps and sales, 20 per department, using simulated business systems and tool contracts. Grading checks the final state, the actions taken, and whether the agent finished the job or escalated it correctly.
Revensi, a company that builds AI agents for businesses, runs the board, which listed 18 models including local ones. The current library is Work 1.3. The repository had 0 GitHub stars on November 5th, 2026, and no outside use was found.
- Made by
- Revensi
- Released
- September 2026
- Kind
- Running a business
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Strongest model tested
-
Muse Spark 1.3 #26 on the capability index - Last model added
- September 2026
- Scores refreshed
- November 12th, 2026
GPT-5.6 Luna leads with 92.7 among models released by September 2026
Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.
The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- Claude Fable 5.2
-
Gemini 4 Carbon - Claude Opus 5.5
- GPT-6.1 Sol
- and 16 more
- 92.7 GPT-5.6 Luna
- 89.6
MiniMax-M3 - 89.2
DeepSeek V4.1 Flash - 88.1 Muse Spark 1.3
- 87.3 Gemini 3.5 Flash-Lite
- 84.6 GLM-5.3-Flash
- 84.0 Gemma 4 E4B
- 79.1 MiMo-V2.5
- 79.1
Nemotron 3.5 Lightning 30B-A3B - 78.6
Tencent Hy3 - 75.7 Mercury 2.5
- 75.6 Claude Haiku 4.5
- 74.8
Mistral Small 4 - 66.2
Inkling-Small - 62.9
Qwen3.8 Flash - 62.4 Qwen3.5-9B
- 46.7 ministral-3:8b
- 31.1 Granite 4.2 8B
- 1 GPT-5.6 Luna CorpBench Work v1.3 92.7
- 2 MiniMax-M3 CorpBench Work v1.3 89.6
- 3 DeepSeek V4.1 Flash CorpBench Work v1.3 89.2
- 4 Muse Spark 1.3 CorpBench Work v1.3 88.1
- 5 Gemini 3.5 Flash-Lite CorpBench Work v1.3 87.3
- 6 GLM-5.3-Flash CorpBench Work v1.3 84.6
- 7 Gemma 4 E4B CorpBench Work v1.3; Q4_K_M; context 32768 84.0
- 8 MiMo-V2.5 CorpBench Work v1.3 79.1
- 9 Nemotron 3.5 Lightning 30B-A3B CorpBench Work v1.3 79.1
- 10 Tencent Hy3 CorpBench Work v1.3 78.6
- 11 Mercury 2.5 CorpBench Work v1.3 75.7
- 12 Claude Haiku 4.5 CorpBench Work v1.3 75.6
- 13 Mistral Small 4 CorpBench Work v1.3 74.8
- 14 Inkling-Small CorpBench Work v1.3 66.2
- 15 Qwen3.8 Flash CorpBench Work v1.3 62.9
- 16 Qwen3.5-9B CorpBench Work v1.3; Q4_K_M; context 32768 62.4
- 17 ministral-3:8b CorpBench Work v1.3; Q4_K_M; context 32768 46.7
- 18 Granite 4.2 8B CorpBench Work v1.3; Q4_K_M; context 32768 31.1
Source: Official CorpBench Work leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5.6 Luna | 92.7 | CorpBench Work v1.3, CorpBench | Jul 2026 | Official leaderboard |
| 2 | MiniMax-M3 | 89.6 | CorpBench Work v1.3, CorpBench | Jun 2026 | Official leaderboard |
| 3 | DeepSeek V4.1 Flash | 89.2 | CorpBench Work v1.3, CorpBench | Sep 2026 | Official leaderboard |
| 4 | Muse Spark 1.3 | 88.1 | CorpBench Work v1.3, CorpBench | Sep 2026 | Official leaderboard |
| 5 | Gemini 3.5 Flash-Lite | 87.3 | CorpBench Work v1.3, CorpBench | Jul 2026 | Official leaderboard |
| 6 | GLM-5.3-Flash | 84.6 | CorpBench Work v1.3, CorpBench | Aug 2026 | Official leaderboard |
| 7 | Gemma 4 E4B | 84.0 | CorpBench Work v1.3; Q4_K_M; context 32768, Ollama | Mar 2026 | Official leaderboard |
| 8 | MiMo-V2.5 | 79.1 | CorpBench Work v1.3, CorpBench | Apr 2026 | Official leaderboard |
| 9 | Nemotron 3.5 Lightning 30B-A3B | 79.1 | CorpBench Work v1.3, CorpBench | Aug 2026 | Official leaderboard |
| 10 | Tencent Hy3 | 78.6 | CorpBench Work v1.3, CorpBench | Jul 2026 | Official leaderboard |
| 11 | Mercury 2.5 | 75.7 | CorpBench Work v1.3, CorpBench | Sep 2026 | Official leaderboard |
| 12 | Claude Haiku 4.5 | 75.6 | CorpBench Work v1.3, CorpBench | Oct 2025 | Official leaderboard |
| 13 | Mistral Small 4 | 74.8 | CorpBench Work v1.3, CorpBench | Mar 2026 | Official leaderboard |
| 14 | Inkling-Small | 66.2 | CorpBench Work v1.3, CorpBench | Jul 2026 | Official leaderboard |
| 15 | Qwen3.8 Flash | 62.9 | CorpBench Work v1.3, CorpBench | Aug 2026 | Official leaderboard |
| 16 | Qwen3.5-9B | 62.4 | CorpBench Work v1.3; Q4_K_M; context 32768, Ollama | Feb 2026 | Official leaderboard |
| 17 | ministral-3:8b | 46.7 | CorpBench Work v1.3; Q4_K_M; context 32768, Ollama | Official leaderboard | |
| 18 | Granite 4.2 8B | 31.1 | CorpBench Work v1.3; Q4_K_M; context 32768, Ollama | Aug 2026 | Official leaderboard |
Performance versus cost
The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.
| Model | Setting | CorpBench Score | Mean cost per task |
|---|---|---|---|
| GPT-5.6 Luna | CorpBench Work v1.3, CorpBench | 92.7 | $0 |
| MiniMax-M3 | CorpBench Work v1.3, CorpBench | 89.6 | $0.01 |
| DeepSeek V4.1 Flash | CorpBench Work v1.3, CorpBench | 89.2 | $0 |
| Muse Spark 1.3 | CorpBench Work v1.3, CorpBench | 88.1 | $0.03 |
| Gemini 3.5 Flash-Lite | CorpBench Work v1.3, CorpBench | 87.3 | $0.01 |
| GLM-5.3-Flash | CorpBench Work v1.3, CorpBench | 84.6 | $0 |
| MiMo-V2.5 | CorpBench Work v1.3, CorpBench | 79.1 | $0 |
| Nemotron 3.5 Lightning 30B-A3B | CorpBench Work v1.3, CorpBench | 79.1 | $0 |
| Tencent Hy3 | CorpBench Work v1.3, CorpBench | 78.6 | $0 |
| Mercury 2.5 | CorpBench Work v1.3, CorpBench | 75.7 | $0 |
| Claude Haiku 4.5 | CorpBench Work v1.3, CorpBench | 75.6 | $0.06 |
| Mistral Small 4 | CorpBench Work v1.3, CorpBench | 74.8 | $0 |
| Inkling-Small | CorpBench Work v1.3, CorpBench | 66.2 | $0.01 |
| Qwen3.8 Flash | CorpBench Work v1.3, CorpBench | 62.9 | $0 |
The best score, 92.7, costs $0.002 a task
Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
Source: Official CorpBench Work leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| GPT-5.6 Luna | CorpBench Work v1.3 | CorpBench | 92.7 | $0.002 | |
| MiniMax-M3 | CorpBench Work v1.3 | CorpBench | 89.6 | $0.005 | |
| DeepSeek V4.1 Flash | CorpBench Work v1.3 | CorpBench | 89.2 | $0.003 | |
| Muse Spark 1.3 | CorpBench Work v1.3 | CorpBench | 88.1 | $0.03 | |
| Gemini 3.5 Flash-Lite | CorpBench Work v1.3 | CorpBench | 87.3 | $0.01 | |
| GLM-5.3-Flash | CorpBench Work v1.3 | CorpBench | 84.6 | $0.003 | |
| Gemma 4 E4B | CorpBench Work v1.3; Q4_K_M; context 32768 | Ollama | 84.0 | ||
| MiMo-V2.5 | CorpBench Work v1.3 | CorpBench | 79.1 | $0.002 | |
| Nemotron 3.5 Lightning 30B-A3B | CorpBench Work v1.3 | CorpBench | 79.1 | $0.004 | |
| Tencent Hy3 | CorpBench Work v1.3 | CorpBench | 78.6 | $0.003 | |
| Mercury 2.5 | CorpBench Work v1.3 | CorpBench | 75.7 | $0.002 | |
| Claude Haiku 4.5 | CorpBench Work v1.3 | CorpBench | 75.6 | $0.06 | |
| Mistral Small 4 | CorpBench Work v1.3 | CorpBench | 74.8 | $0.001 | |
| Inkling-Small | CorpBench Work v1.3 | CorpBench | 66.2 | $0.008 | |
| Qwen3.8 Flash | CorpBench Work v1.3 | CorpBench | 62.9 | $0.004 | |
| Qwen3.5-9B | CorpBench Work v1.3; Q4_K_M; context 32768 | Ollama | 62.4 | ||
| Granite 4.2 8B | CorpBench Work v1.3; Q4_K_M; context 32768 | Ollama | 31.1 |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official CorpBench Work leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official CorpBench Work leaderboard
No lab model card reports it yet
We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.
More in Economic work
- JobBench In use 130 tasks from 35 occupations workers most want delegated. Agent works through messy reference files, graded by fact-anchored binary rubrics. GLM-5.3 61.4%
- GDPval In use Real work tasks from 44 occupations with reference files. Experts blindly compare model deliverables with expert work. Metric is win-or-tie rate.
Seed2.1 Pro 87.9% - GDPval-AA v2.1 In use Agents with shell and web access complete 220 GDPval gold tasks from 44 occupations. Blind pairwise comparisons give Elo ratings. Claude Opus 5.5 1,846
- AA-Briefcase v1.1 In use Version 1.1 of Artificial Analysis's multi-hour office work evaluation, with Elo ratings fitted by Crowd-BT. Claude Opus 5.5 1,822
- CoWorkBench In use Agent completes long-horizon productivity tasks across computer science, finance, law and medicine. Qwen in-house scoring (method unpublished). Claude Fable 5 75.9%
- OfficeQA In use Grounded reasoning over a large corpus of U.S. Treasury Bulletin documents: locate tables and compute precise numeric answers. Claude Opus 4.7 86.3%