MortgageTax
Reading tax certificate images and extracting the figures. Run by Vals AI.
As of 19 Sept 2026, Claude Opus 5 leads MortgageTax on BenchLeader with 72.1%, ahead of Claude Fable 5.1 at 70.8%, across 93 model configurations with a published result.
- Published by
- Vals AI
- Category
- Multimodal
- Index weight
- Reference only
- Models
- 93
- Data as of
- 19 Sept 2026
Vals AI (vals.ai).
What the test looks like
Scanned tax certificates must be read and their figures extracted correctly, a document-understanding task with exact answers.
How it is scored
Accuracy, run by Vals AI.
What to keep in mind
Document OCR and reasoning together; only vision models are listed.
- 1Claude Opus 572.1%
- 2Claude Fable 5.170.8%
- 3Claude Opus 4.770.3%
- 4Claude Sonnet 570.0%
- 5Claude Opus 4.869.9%
- 6Gemini 3.1 Pro (high)69.4%
- 7Gemini 3 Pro (high)69.1%
- 8Claude Fable 568.9%
- 9Gemini 2.5 Pro68.9%
- 10GPT-5.5 (xhigh)68.8%
- 11Gemini 3 Flash (high)68.7%
- 12Claude Opus 4.568.7%
- 13Claude 3.7 Sonnet68.7%
- 14Gemini 3.5 Flash Lite (high)68.7%
- 15Claude Opus 4.6 (thinking)68.5%
93 of 93
| # | ||||
|---|---|---|---|---|
| 1 |  Claude Opus 5Anthropic | 72.1% | 66.7 | 2026-09-01 |
| 2 | Claude Fable 5.1Anthropic | 70.8% | 64.5 | 2026-09-01 |
| 3 | Claude Opus 4.7Anthropic | 70.3% | 64.5 | 2026-09-01 |
| 4 | Claude Sonnet 5Anthropic | 70.0% | 57.2 | 2026-09-01 |
| 5 | Claude Opus 4.8Anthropic | 69.9% | 61.9 | 2026-09-01 |
| 6 |  Gemini 3.1 ProhighGoogle | 69.4% | 60.7 | 2026-09-01 |
| 7 | Gemini 3 ProhighGoogle | 69.1% | 61.1 | 2026-09-01 |
| 8 | Claude Fable 5Anthropic | 68.9% | 68.3 | 2026-09-01 |
| 9 | Gemini 2.5 ProGoogle | 68.9% | 54.5 | 2026-09-01 |
| 10 |  GPT-5.5xhighOpenAI | 68.8% | 67.6 | 2026-09-01 |
| 11 | Gemini 3 FlashhighGoogle | 68.7% | 58.7 | 2026-09-01 |
| 12 | Claude Opus 4.5Anthropic | 68.7% | 58.2 | 2026-09-01 |
| 13 | Claude 3.7 SonnetAnthropic | 68.7% | 50.2 | 2026-09-01 |
| 14 | Gemini 3.5 Flash LitehighGoogle | 68.7% | 44.3 | 2026-09-01 |
| 15 | Claude Opus 4.6thinkingAnthropic | 68.5% | 62.5 | 2026-09-01 |
| 16 |  MiniMax-M3MiniMaxopen ↗ | 68.4% | 57.5 | 2026-09-01 |
| 17 | GPT-5.4xhighOpenAI | 68.3% | 65.4 | 2026-09-01 |
| 18 |  Qwen3.6 27BAlibaba | 68.3% | 44.7 | 2026-09-01 |
| 19 | Gemini 3.5 FlashhighGoogle | 68.1% | 63.6 | 2026-09-01 |
| 20 | Gemini 3.1 Flash LitehighGoogle | 68.0% | 49.1 | 2026-09-01 |
| 21 | Qwen3.6 PlusAlibabaopen | 68.0% | 58.5 | 2026-09-01 |
| 22 | Claude Sonnet 4.6Anthropic | 67.7% | 59.0 | 2026-09-01 |
| 23 | Claude Opus 4.5thinkingAnthropic | 67.7% | 60.8 | 2026-09-01 |
| 24 | Gemini 3.6 FlashhighGoogle | 67.6% | 61.7 | 2026-09-01 |
| 25 | Qwen3.5 FlashAlibaba | 67.4% | 52.5 | 2026-09-01 |
| 26 | GPT-5.6 TerraxhighOpenAI | 67.3% | 64.3 | 2026-09-01 |
| 27 | GPT-5.6 SolmaxOpenAI | 67.3% | 68.8 | 2026-09-01 |
| 28 | GPT-5.6 LunamaxOpenAI | 67.3% | 60.2 | 2026-09-01 |
| 29 | GPT-5.2xhighOpenAI | 67.1% | 62.0 | 2026-09-01 |
| 30 | GPT-5 minihighOpenAI | 66.9% | 55.3 | 2026-09-01 |
| 31 | Claude 3.7 SonnetthinkingAnthropic | 66.8% | 51.7 | 2026-09-01 |
| 32 | Gemini 3.7 FlashhighGoogle | 66.7% | 64.6 | 2026-09-01 |
| 33 |  Kimi K2.5thinkingMoonshot AIopen | 66.5% | 58.5 | 2026-09-01 |
| 34 | Kimi K3Moonshot AIopen ↗ | 66.3% | 64.2 | 2026-09-01 |
| 35 | Qwen3.7 PlusAlibaba | 66.2% | 59.9 | 2026-09-01 |
| 36 |  Muse Spark 1.1xhighMeta | 66.1% | 61.8 | 2026-09-01 |
| 37 | GPT-4.1highOpenAI | 65.9% | 46.6 | 2026-09-01 |
| 38 | Kimi K2.6Moonshot AIopen ↗ | 65.8% | 60.5 | 2026-09-01 |
| 39 | o3highOpenAI | 65.7% | 53.0 | 2026-09-01 |
| 40 | GPT-4.1 minihighOpenAI | 65.5% | 46.3 | 2026-09-01 |
| 41 | GPT-5highOpenAI | 65.5% | 58.6 | 2026-09-01 |
| 42 | Muse Spark 1.2xhighMeta | 65.4% | 64.1 | 2026-09-01 |
| 43 | Gemini 3.8 FlashhighGoogle | 65.3% | 64.5 | 2026-09-01 |
| 44 | Qwen3.8 27BxhighAlibabaopen ↗ | 64.9% | 59.2 | 2026-09-01 |
| 45 | o4-minihighOpenAI | 64.8% | 50.8 | 2026-09-01 |
| 46 |  Grok 4.6highxAI | 64.2% | 64.5 | 2026-09-01 |
| 47 | Claude 3.5 SonnetAnthropic | 64.1% | 46.6 | 2026-09-01 |
| 48 | Qwen3 8maxAlibaba | 64.0% | 66.5 | 2026-09-01 |
| 49 | Claude Sonnet 4.5thinkingAnthropic | 64.0% | 53.4 | 2026-09-01 |
| 50 | GPT-5.4 minixhighOpenAI | 63.5% | 55.7 | 2026-09-01 |
| 51 | Gemini 2.5 Flash 09 2025Google | 62.6% | 52.1 | 2026-09-01 |
| 52 | Claude Sonnet 4Anthropic | 62.5% | 49.7 | 2026-09-01 |
| 53 | Inkling SmallThinking Machinesopen ↗ | 62.3% | 56.3 | 2026-09-01 |
| 54 | Claude Haiku 4.5thinkingAnthropic | 62.2% | 47.5 | 2026-09-01 |
| 55 |  Magistral Small 1.2Mistral AIopen | 62.1% | 40.9 | 2026-09-01 |
| 56 | Gemini 2.5 Flash 09 2025thinkingGoogle | 61.9% | 51.6 | 2026-09-01 |
| 57 | Grok 4.5highxAI | 61.8% | 61.0 | 2026-09-01 |
| 58 | GPT-5.1highOpenAI | 61.4% | 58.7 | 2026-09-01 |
| 59 | Gemma 4 31BhighGoogleopen ↗ | 61.4% | – | 2026-09-01 |
| 60 | Mistral Small 3.1Mistral AI | 61.2% | 34.6 | 2026-09-01 |
Cite as: BenchLeader, “MortgageTax leaderboard”, https://www.benchleader.com/benchmarks/vals_mortgage_tax, data as of 19 Sept 2026.
MortgageTax: questions
- What does MortgageTax measure?
- Scanned tax certificates must be read and their figures extracted correctly, a document-understanding task with exact answers. Scores are reported in percent of tasks solved; higher is better.
- Which AI model leads MortgageTax?
- Claude Opus 5 leads MortgageTax with 72.1% as of 19 Sept 2026, ahead of Claude Fable 5.1 at 70.8%.
- How many models have MortgageTax results?
- 93 model configurations have a MortgageTax result on BenchLeader, all taken from Vals AI.
- Who runs MortgageTax and how often is it updated?
- MortgageTax is published by Vals AI. BenchLeader re-reads the published results every morning and records the date each result was published.
- Does MortgageTax count toward the BenchLeader Index?
- No. MortgageTax is shown for reference but left out of the composite index.