MM-FinEval
Models forecast financial outcomes from earnings call transcripts, slides, and audio. The authors compare 19 models using directional accuracy and absolute error.
- Made by
- Northwestern University and collaborators
- Released
- September 2026
- Kind
- Finance
- Level
- University
- Scored
- Checked automatically
- Answers
- Prediction
- Tasks
- All public
- Measures
- Capability
- Subject
- Economics and business
- Language
- English
- Interaction
- Single turn
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Audio and speech, Law and finance
- Strongest model tested
- GPT-5.5 #21 on the capability index
- Last model added
- May 2026
- Scores refreshed
- November 20th, 2026
GPT-5.5 leads with 88% among models released by May 2026 GPT-5.5 leads with 88% among models released by May 2026
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper gives no overall directional accuracy, so this is the plain mean of its 12 task accuracies. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in May 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 88.0 GPT-5.5
- 87.9 Claude Sonnet 4.6
- 87.3 Claude Opus 4.8
- 85.8 GPT-5.4
- 83.9
Qwen3.5-35B-A3B
- 83.8
Qwen3.5-27B
- 83.8
Qwen3.6 27B
- 83.6 Mis3-14B
- 83.4
Gemma 4 E2B
- 83.0 Gm3n
- 81.5 Mis3-8B
- 81.0
Qwen 3.6 35B-A3B
- 80.7
Qwen3.5-9B
- 73.2
Gemma 4 E4B
- 69.3
Qwen3.5-4B
- 49.6 Moss
- 44.3 Mis3-3B
- 34.0
Voxtral Small
- 28.2
Qwen3.5-2B
- 1 GPT-5.5 Image-Text 88%
- 2 Claude Sonnet 4.6 Image-Text 87.9%
- 3 Claude Opus 4.8 Image-Text 87.3%
- 4 GPT-5.4 Image-Text 85.8%
- 5
Qwen3.5-35B-A3B Image-Text 83.9%
- 6
Qwen3.5-27B Image-Text 83.8%
- 7
Qwen3.6 27B Image-Text 83.8%
- 8 Mis3-14B Image-Text 83.6%
- 9
Gemma 4 E2B Any-to-Any 83.4%
- 10 Gm3n Image-Text 83%
- 11 Mis3-8B Image-Text 81.5%
- 12
Qwen 3.6 35B-A3B Image-Text 81%
- 13
Qwen3.5-9B Image-Text 80.7%
- 14
Gemma 4 E4B Any-to-Any 73.2%
- 15
Qwen3.5-4B Image-Text 69.3%
- 16 Moss Audio-Text 49.6%
- 17 Mis3-3B Image-Text 44.3%
- 18
Voxtral Small Audio-Text 34%
- 19
Qwen3.5-2B Image-Text 28.2%
Source: The MM-FinEval paper
- 1 GPT-5.5 Image-Text 88%
- 2 Claude Sonnet 4.6 Image-Text 87.9%
- 3 Claude Opus 4.8 Image-Text 87.3%
- 4 GPT-5.4 Image-Text 85.8%
- 5
Qwen3.5-35B-A3B Image-Text 83.9%
- 6
Qwen3.5-27B Image-Text 83.8%
- 7
Qwen3.6 27B Image-Text 83.8%
- 8
Gemma 4 E2B Any-to-Any 83.4%
- 9
Qwen 3.6 35B-A3B Image-Text 81%
- 10
Qwen3.5-9B Image-Text 80.7%
- 11
Gemma 4 E4B Any-to-Any 73.2%
- 12
Qwen3.5-4B Image-Text 69.3%
- 13
Voxtral Small Audio-Text 34%
- 14
Qwen3.5-2B Image-Text 28.2%
Source: The MM-FinEval paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5.5 | 88% | Image-Text | Apr 2026 | Official leaderboard |
| 2 | Claude Sonnet 4.6 | 87.9% | Image-Text | Feb 2026 | Official leaderboard |
| 3 | Claude Opus 4.8 | 87.3% | Image-Text | May 2026 | Official leaderboard |
| 4 | GPT-5.4 | 85.8% | Image-Text | Mar 2026 | Official leaderboard |
| 5 | | 83.9% | Image-Text | Feb 2026 | Official leaderboard |
| 6 | | 83.8% | Image-Text | Feb 2026 | Official leaderboard |
| 7 | | 83.8% | Image-Text | Apr 2026 | Official leaderboard |
| 8 | Mis3-14B | 83.6% | Image-Text | Official leaderboard | |
| 9 | | 83.4% | Any-to-Any | Mar 2026 | Official leaderboard |
| 10 | Gm3n | 83% | Image-Text | Official leaderboard | |
| 11 | Mis3-8B | 81.5% | Image-Text | Official leaderboard | |
| 12 | | 81% | Image-Text | Apr 2026 | Official leaderboard |
| 13 | | 80.7% | Image-Text | Feb 2026 | Official leaderboard |
| 14 | | 73.2% | Any-to-Any | Mar 2026 | Official leaderboard |
| 15 | | 69.3% | Image-Text | Feb 2026 | Official leaderboard |
| 16 | Moss | 49.6% | Audio-Text | Official leaderboard | |
| 17 | Mis3-3B | 44.3% | Image-Text | Official leaderboard | |
| 18 | | 34% | Audio-Text | Jul 2025 | Official leaderboard |
| 19 | | 28.2% | Image-Text | Feb 2026 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5.5 | 88% | Image-Text | Apr 2026 | Official leaderboard |
| 2 | Claude Sonnet 4.6 | 87.9% | Image-Text | Feb 2026 | Official leaderboard |
| 3 | Claude Opus 4.8 | 87.3% | Image-Text | May 2026 | Official leaderboard |
| 4 | GPT-5.4 | 85.8% | Image-Text | Mar 2026 | Official leaderboard |
| 5 | | 83.9% | Image-Text | Feb 2026 | Official leaderboard |
| 6 | | 83.8% | Image-Text | Feb 2026 | Official leaderboard |
| 7 | | 83.8% | Image-Text | Apr 2026 | Official leaderboard |
| 8 | | 83.4% | Any-to-Any | Mar 2026 | Official leaderboard |
| 9 | | 81% | Image-Text | Apr 2026 | Official leaderboard |
| 10 | | 80.7% | Image-Text | Feb 2026 | Official leaderboard |
| 11 | | 73.2% | Any-to-Any | Mar 2026 | Official leaderboard |
| 12 | | 69.3% | Image-Text | Feb 2026 | Official leaderboard |
| 13 | | 34% | Audio-Text | Jul 2025 | Official leaderboard |
| 14 | | 28.2% | Image-Text | Feb 2026 | Official leaderboard |
Rankings by part
We read 24 parts from the benchmark’s paper in 2 groups: directional accuracy and mean absolute error. Each ranks models on one part of the test. Claude Sonnet 4.6 leads 7 of them, and 5 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
car
14 models we track have a score here.
- 1
Gemma 4 E2B ↑7 55.2%
- 2 Claude Sonnet 4.6 53.7%
- 3 GPT-5.5 ↓2 53.5%
- 4
Gemma 4 E4B ↑7 53.2%
- 5
Qwen3.5-35B-A3B 53.1%
- 6 GPT-5.4 ↓2 51.8%
- 7
Qwen3.6 27B 51.6%
- 8
Qwen 3.6 35B-A3B ↑1 51.1%
- 9
Qwen3.5-27B ↓3 50.6%
- 10
Qwen3.5-9B 48.1%
ret_eom
14 models we track have a score here.
- 1 Claude Opus 4.8 ↑2 72.4%
- 2 Claude Sonnet 4.6 71.9%
- 3 GPT-5.5 ↓2 71.5%
- 4
Gemma 4 E2B ↑4 60.6%
- 5
Gemma 4 E4B ↑6 60.1%
- 6 GPT-5.4 ↓2 59.5%
- 7
Qwen 3.6 35B-A3B ↑2 57.1%
- 8
Qwen3.6 27B ↓1 56.8%
- 9
Qwen3.5-27B ↓3 56.2%
- 10
Qwen3.5-35B-A3B ↓5 55.5%
ret_l5m
14 models we track have a score here.
- 1 Claude Opus 4.8 ↑2 67.5%
- 2 Claude Sonnet 4.6 67%
- 3 GPT-5.5 ↓2 66%
- 4
Gemma 4 E4B ↑7 61.1%
- 5 GPT-5.4 ↓1 59.2%
- 6
Gemma 4 E2B ↑2 59.1%
- 7
Qwen3.5-27B ↓1 56.7%
- 8
Qwen3.6 27B ↓1 56.1%
- 9
Qwen3.5-35B-A3B ↓4 55.4%
- 10
Qwen 3.6 35B-A3B ↓1 54%
ret_ldm
14 models we track have a score here.
- 1 Claude Sonnet 4.6 ↑1 68%
- 2 Claude Opus 4.8 ↑1 66.5%
- 3 GPT-5.5 ↓2 64.5%
- 4
Gemma 4 E2B ↑4 62.6%
- 5
Gemma 4 E4B ↑6 62.1%
- 6 GPT-5.4 ↓2 59.2%
- 7
Qwen 3.6 35B-A3B ↑2 56.6%
- 8
Qwen3.5-27B ↓2 55.2%
- 9
Qwen3.5-35B-A3B ↓4 54.4%
- 10
Qwen3.6 27B ↓3 53.7%
fut_3d
14 models we track have a score here.
- 1 Claude Sonnet 4.6 ↑1 100%
- 2 GPT-5.5 ↓1 100%
- 3 Claude Opus 4.8 100%
- 4 GPT-5.4 99.8%
- 5
Gemma 4 E4B ↑6 99.5%
- 6
Qwen3.5-35B-A3B ↓1 99.4%
- 7
Qwen 3.6 35B-A3B ↑2 99.3%
- 8
Gemma 4 E2B 98%
- 9
Qwen3.5-27B ↓3 96.6%
- 10
Qwen3.6 27B ↓3 96.3%
fut_7d
14 models we track have a score here.
- 1 GPT-5.4 ↑3 100%
- 2 GPT-5.5 ↓1 100%
- 3 Claude Opus 4.8 100%
- 4
Gemma 4 E4B ↑7 99.5%
- 5
Qwen3.5-35B-A3B 99.2%
- 6 Claude Sonnet 4.6 ↓4 99%
- 7
Qwen 3.6 35B-A3B ↑2 98.7%
- 8
Gemma 4 E2B 98.5%
- 9
Qwen3.6 27B ↓2 97.1%
- 10
Qwen3.5-4B ↑2 95.9%
fut_15d
14 models we track have a score here.
- 1 GPT-5.4 ↑3 100%
- 2 GPT-5.5 ↓1 100%
- 3 Claude Opus 4.8 100%
- 4
Gemma 4 E4B ↑7 99.3%
- 5
Qwen3.5-35B-A3B 99.3%
- 6
Qwen 3.6 35B-A3B ↑3 99.2%
- 7
Gemma 4 E2B ↑1 98.5%
- 8
Qwen3.5-27B ↓2 98.5%
- 9
Qwen3.6 27B ↓2 98.3%
- 10 Claude Sonnet 4.6 ↓8 97.5%
fut_30d
14 models we track have a score here.
- 1 GPT-5.4 ↑3 100%
- 2 GPT-5.5 ↓1 100%
- 3 Claude Opus 4.8 100%
- 4
Gemma 4 E4B ↑7 99.2%
- 5
Qwen3.5-35B-A3B 98.8%
- 6
Qwen 3.6 35B-A3B ↑3 98.6%
- 7
Qwen3.6 27B 98.1%
- 8 Claude Sonnet 4.6 ↓6 98%
- 9
Gemma 4 E2B ↓1 98%
- 10
Qwen3.5-27B ↓4 97.9%
past_3d
14 models we track have a score here.
- 1 Claude Sonnet 4.6 ↑1 100%
- 2 GPT-5.4 ↑2 100%
- 3 GPT-5.5 ↓2 100%
- 4 Claude Opus 4.8 ↓1 100%
- 5
Qwen3.5-27B ↑1 99.2%
- 6
Qwen3.6 27B ↑1 98.9%
- 7
Qwen3.5-35B-A3B ↓2 95.5%
- 8
Qwen3.5-9B ↑2 95.3%
- 9
Gemma 4 E2B ↓1 94.6%
- 10
Qwen 3.6 35B-A3B ↓1 84.2%
past_7d
14 models we track have a score here.
- 1 Claude Sonnet 4.6 ↑1 100%
- 2 GPT-5.4 ↑2 100%
- 3 GPT-5.5 ↓2 100%
- 4 Claude Opus 4.8 ↓1 100%
- 5
Qwen3.5-27B ↑1 99.7%
- 6
Qwen3.6 27B ↑1 99.3%
- 7
Qwen3.5-35B-A3B ↓2 98%
- 8
Qwen3.5-9B ↑2 95.1%
- 9
Gemma 4 E2B ↓1 91.6%
- 10
Qwen 3.6 35B-A3B ↓1 89.2%
past_15d
14 models we track have a score here.
- 1 Claude Sonnet 4.6 ↑1 100%
- 2 GPT-5.4 ↑2 100%
- 3 GPT-5.5 ↓2 100%
- 4 Claude Opus 4.8 ↓1 100%
- 5
Qwen3.5-27B ↑1 99.7%
- 6
Qwen3.6 27B ↑1 99.5%
- 7
Qwen3.5-35B-A3B ↓2 99.1%
- 8
Qwen3.5-9B ↑2 95.1%
- 9
Gemma 4 E2B ↓1 92.6%
- 10
Qwen 3.6 35B-A3B ↓1 89.6%
past_30d
14 models we track have a score here.
- 1 Claude Sonnet 4.6 ↑1 100%
- 2 GPT-5.4 ↑2 100%
- 3 GPT-5.5 ↓2 100%
- 4 Claude Opus 4.8 ↓1 100%
- 5
Qwen3.5-27B ↑1 99.6%
- 6
Qwen3.5-35B-A3B ↓1 99.5%
- 7
Qwen3.6 27B 99.3%
- 8
Qwen3.5-9B ↑2 94.7%
- 9
Qwen 3.6 35B-A3B 94.7%
- 10
Gemma 4 E2B ↓2 91.6%
car
14 models we track have a score here.
- 1
Gemma 4 E2B ↑7 0.04
- 2
Voxtral Small ↑11 0.04
- 3
Gemma 4 E4B ↑8 0.04
- 4 Claude Opus 4.8 ↓1 0.05
- 5
Qwen3.5-35B-A3B 0.05
- 6
Qwen 3.6 35B-A3B ↑3 0.05
- 7 Claude Sonnet 4.6 ↓5 0.05
- 8 GPT-5.4 ↓4 0.05
- 9
Qwen3.5-4B ↑3 0.05
- 10
Qwen3.5-9B 0.05
ret_eom
14 models we track have a score here.
- 1 GPT-5.5 0.01
- 2 Claude Opus 4.8 ↑1 0.02
- 3 Claude Sonnet 4.6 ↓1 0.02
- 4 GPT-5.4 0.02
- 5
Qwen 3.6 35B-A3B ↑4 0.02
- 6
Qwen3.5-4B ↑6 0.02
- 7
Qwen3.5-35B-A3B ↓2 0.02
- 8
Qwen3.5-9B ↑2 0.02
- 9
Voxtral Small ↑4 0.02
- 10
Qwen3.6 27B ↓3 0.02
ret_l5m
14 models we track have a score here.
- 1 GPT-5.5 0.02
- 2 GPT-5.4 ↑2 0.02
- 3 Claude Sonnet 4.6 ↓1 0.02
- 4
Qwen3.5-35B-A3B ↑1 0.02
- 5
Qwen 3.6 35B-A3B ↑4 0.02
- 6 Claude Opus 4.8 ↓3 0.02
- 7
Qwen3.5-4B ↑5 0.02
- 8
Qwen3.6 27B ↓1 0.02
- 9
Qwen3.5-9B ↑1 0.02
- 10
Voxtral Small ↑3 0.02
ret_ldm
14 models we track have a score here.
- 1 GPT-5.5 0.02
- 2 GPT-5.4 ↑2 0.02
- 3
Qwen 3.6 35B-A3B ↑6 0.02
- 4 Claude Sonnet 4.6 ↓2 0.02
- 5
Qwen3.5-35B-A3B 0.02
- 6 Claude Opus 4.8 ↓3 0.02
- 7
Qwen3.6 27B 0.02
- 8
Qwen3.5-4B ↑4 0.02
- 9
Qwen3.5-9B ↑1 0.02
- 10
Qwen3.5-27B ↓4 0.02
fut_3d
14 models we track have a score here.
- 1
Gemma 4 E4B ↑10 0.01
- 2
Gemma 4 E2B ↑6 0.01
- 3 Claude Opus 4.8 0.01
- 4 Claude Sonnet 4.6 ↓2 0.01
- 5
Qwen 3.6 35B-A3B ↑4 0.02
- 6
Qwen3.5-35B-A3B ↓1 0.02
- 7
Voxtral Small ↑6 0.02
- 8
Qwen3.6 27B ↓1 0.02
- 9
Qwen3.5-4B ↑3 0.02
- 10 GPT-5.5 ↓9 0.02
fut_7d
14 models we track have a score here.
- 1
Gemma 4 E4B ↑10 0.01
- 2
Gemma 4 E2B ↑6 0.01
- 3 Claude Opus 4.8 0.01
- 4 Claude Sonnet 4.6 ↓2 0.01
- 5
Qwen 3.6 35B-A3B ↑4 0.02
- 6
Qwen3.5-35B-A3B ↓1 0.02
- 7
Qwen3.5-4B ↑5 0.02
- 8
Qwen3.6 27B ↓1 0.02
- 9
Voxtral Small ↑4 0.02
- 10
Qwen3.5-27B ↓4 0.02
fut_15d
14 models we track have a score here.
- 1
Gemma 4 E4B ↑10 0.01
- 2
Gemma 4 E2B ↑6 0.01
- 3 Claude Opus 4.8 0.01
- 4 Claude Sonnet 4.6 ↓2 0.02
- 5
Qwen3.5-35B-A3B 0.02
- 6
Voxtral Small ↑7 0.02
- 7
Qwen3.5-4B ↑5 0.02
- 8
Qwen3.6 27B ↓1 0.02
- 9
Qwen3.5-27B ↓3 0.02
- 10
Qwen3.5-2B ↑4 0.02
fut_30d
14 models we track have a score here.
- 1
Gemma 4 E4B ↑10 0.01
- 2
Gemma 4 E2B ↑6 0.01
- 3
Qwen3.5-35B-A3B ↑2 0.02
- 4 Claude Opus 4.8 ↓1 0.02
- 5
Voxtral Small ↑8 0.02
- 6
Qwen3.5-4B ↑6 0.02
- 7
Qwen3.5-27B ↓1 0.02
- 8
Qwen3.5-2B ↑6 0.02
- 9 GPT-5.5 ↓8 0.03
- 10
Qwen3.5-9B 0.03
past_3d
14 models we track have a score here.
- 1 Claude Sonnet 4.6 ↑1 0.01
- 2 Claude Opus 4.8 ↑1 0.01
- 3 GPT-5.5 ↓2 0.01
- 4 GPT-5.4 0.01
- 5
Gemma 4 E4B ↑6 0.02
- 6
Gemma 4 E2B ↑2 0.02
- 7
Qwen3.5-35B-A3B ↓2 0.02
- 8
Qwen3.6 27B ↓1 0.02
- 9
Qwen3.5-2B ↑5 0.02
- 10
Qwen3.5-4B ↑2 0.02
past_7d
14 models we track have a score here.
- 1 Claude Opus 4.8 ↑2 0.01
- 2
Qwen3.6 27B ↑5 0.01
- 3
Gemma 4 E4B ↑8 0.01
- 4 GPT-5.4 0.01
- 5 Claude Sonnet 4.6 ↓3 0.01
- 6
Qwen3.5-35B-A3B ↓1 0.02
- 7 GPT-5.5 ↓6 0.02
- 8
Qwen3.5-2B ↑6 0.02
- 9
Qwen3.5-4B ↑3 0.02
- 10
Gemma 4 E2B ↓2 0.02
past_15d
14 models we track have a score here.
- 1 Claude Opus 4.8 ↑2 0.01
- 2
Gemma 4 E4B ↑9 0.01
- 3 GPT-5.4 ↑1 0.01
- 4
Qwen3.5-35B-A3B ↑1 0.01
- 5
Qwen3.6 27B ↑2 0.02
- 6 GPT-5.5 ↓5 0.02
- 7
Qwen3.5-4B ↑5 0.02
- 8
Qwen3.5-2B ↑6 0.02
- 9
Qwen3.5-27B ↓3 0.02
- 10
Voxtral Small ↑3 0.02
past_30d
14 models we track have a score here.
- 1
Gemma 4 E4B ↑10 0.01
- 2
Qwen3.5-35B-A3B ↑3 0.02
- 3
Qwen3.5-4B ↑9 0.02
- 4
Qwen3.5-2B ↑10 0.02
- 5
Voxtral Small ↑8 0.02
- 6 GPT-5.5 ↓5 0.02
- 7 GPT-5.4 ↓3 0.02
- 8
Qwen3.5-27B ↓2 0.04
- 9
Qwen3.5-9B ↑1 0.04
- 10 Claude Opus 4.8 ↓7 0.06
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The MM-FinEval paper
Claude Sonnet 4.6 on the official leaderboard passed 85% in February 2026
Models released before it came out already scored that high. All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official MM-FinEval leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Law, finance and medicine
- MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy.
Qwen 3.8 Max 80.4%
- HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
- HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
- FinSearchComp In use Agents search the web to fetch live financial data, look up history, and run multi-step investigations, across 635 expert-written questions. Scored by answer accuracy.
Ling-3.1-flash 81.7%
- KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. o1-preview 92.7%
- Harvey LAB-AA v1.1 In use Artificial Analysis's run of Harvey's Legal Agent Benchmark on Harvey's newer private set of 120 tasks across 24 practice areas. A deliverable with a material hallucination scores zero, and the headline is the share of judges who find every rubric criterion passed. GPT-6.2 Astra 15%
Data audit
All 8 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
- Pass Independent runs Scores from the benchmark’s paper (19 entries).
- Pass Models scored 14 models we track, plus 5 other entries on its leaderboard.
- Note Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. That’s expected for a saturated test.
- Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is GPT-5.5 (#21).
- Note Headroom Saturated. The best is 88%, past 85% since February 2026.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 14 rows for models we track and 5 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.