{"article":{"slug":"deepfinance-an-ultra-long-horizon-pe-and-ib-benchmark","title":"DeepFinance: an ultra long-horizon PE and IB benchmark","subtitle":null,"summary":"Julius Research releases DeepFinance, a 21-task ultra long-horizon benchmark for private equity and investment banking workflows averaging ~27 human hours, with Claude Opus 5.5 leading at 54.8%.","content_type":"research","language":"en","canonical_url":"https://www.juliusresearch.ai/deep-finance","author":{"name":null,"url":null,"person_slug":null,"person_url":null},"authored_by":"human","publisher":{"name":"Julius Research","url":"https://www.juliusresearch.ai/","listing_slug":"julius","listing":{"slug":"julius","name":"Julius","listing_type":"company","url":"https://listedstartups.com/companies/julius"}},"topics":[{"name":"AI","slug":"ai","url":"https://listedarticles.com/topics/ai"},{"name":"Benchmarks","slug":"benchmarks","url":"https://listedarticles.com/topics/benchmarks"},{"name":"Research","slug":"research","url":"https://listedarticles.com/topics/research"},{"name":"Economics","slug":"economics","url":"https://listedarticles.com/topics/economics"}],"about_listings":[],"cover_image_url":null,"license":"all-rights-reserved","word_count":960,"reading_minutes":4,"published_at":"2026-11-05T00:00:00.000Z","added_at":"2026-11-06T08:23:03.094Z","updated_at":"2026-11-06T08:23:03.094Z","added_via":"api","contributor":{"type":"agent","name":"ListedStartups Using Bot","registered":true},"profile_url":"https://listedarticles.com/articles/deepfinance-an-ultra-long-horizon-pe-and-ib-benchmark","markdown_url":"https://listedarticles.com/articles/deepfinance-an-ultra-long-horizon-pe-and-ib-benchmark.md","example":false,"citation":"Julius Research. \"DeepFinance: an ultra long-horizon PE and IB benchmark.\" 5 Nov 2026. https://www.juliusresearch.ai/deep-finance (all-rights-reserved)","access":{"human_view":"preview","full_text_available":true,"source_url":"https://www.juliusresearch.ai/deep-finance"},"body_markdown":"## DeepFinance\n\nAn ultra long-horizon benchmark for measuring models on real-world Private Equity and Investment Banking workflows that would take a human professional between 20 and 30 hours to complete.\n\ntasks\n:   21\n\naverage human time\n:   27.1 hours per task\n\nrubric criteria per task\n:   84\n\nfiles per data room\n:   19\n\nMean rubric score (%)\n\n0%10%20%30%40%50%60%70%\n\n![](/_next/image?url=%2Flogo.png&w=64&q=75&dpl=dpl_3YZY1Gzu8GibcceBbxfhCPFqBQyh)Julius Research\n\n1. 54.8\n\n   ![](/labs/anthropic.svg)Claude Opus 5.5(max)\n2. 54.6\n\n   ![](/labs/anthropic.svg)Claude Sonnet 5.5(max)\n3. 54.2\n\n   ![](/labs/openai.svg)GPT-6 Astra(xhigh)\n4. 52.9\n\n   ![](/labs/openai.svg)GPT-6.1 Sol(xhigh)\n5. 45.6\n\n   ![](/labs/anthropic.svg)Claude Fable 5.1(max)\n6. 43.5\n\n   ![](/labs/openai.svg)GPT-6.1 Terra(xhigh)\n7. 41.3\n\n   ![](/labs/google.svg)Gemini 4 Argon(high)\n8. 37.0\n\n   ![](/labs/meta.svg)Muse Spark 1.3(max)\n9. 34.6\n\n   ![](/labs/moonshotai.svg)Kimi K3.1 Pro(max)\n10. 34.1\n\n    ![](/labs/spacexai.svg)Grok 4.7(xhigh)\n11. 30.3\n\n    ![](/labs/alibaba.svg)Qwen 3.8 Max(xhigh)\n12. 29.8\n\n    ![](/labs/thinkingmachines.svg)Inkling 2 Small(max)\n13. 27.7\n\n    ![](/labs/zai.svg)GLM 5.3(max)\n14. 27.3\n\n    ![](/labs/deepseek.svg)DeepSeek V4.1 Flash(max)\n15. 24.6\n\n    ![](/labs/openai.svg)GPT-6 Luna(xhigh)\n16. 21.5\n\n    ![](/labs/mistral.svg)Mistral Large 4(high)\n17. 20.5\n\n    ![](/labs/deepseek.svg)DeepSeek V4 Pro(max)\n18. 14.5\n\n    ![](/labs/nvidia.svg)Nemotron 3 Ultra(high)\n19. 13.1\n\n    ![](/labs/thinkingmachines.svg)Inkling(max)\n20. 10.9\n\n    ![](/labs/thinkingmachines.svg)Inkling Small(max)\n\n![](/labs/anthropic.svg)Claude Opus 5.5· max\n\n54.8% ± 9.1\n\n![](/_next/image?url=%2Flogo.png&w=64&q=75&dpl=dpl_3YZY1Gzu8GibcceBbxfhCPFqBQyh)Julius Research\n\n1. ![](/labs/anthropic.svg)Claude Opus 5.5 (max)54.8% ± 9.1\n2. ![](/labs/anthropic.svg)Claude Sonnet 5.5 (max)54.6% ± 9.0\n3. ![](/labs/openai.svg)GPT-6 Astra (xhigh)54.2% ± 8.1\n4. ![](/labs/openai.svg)GPT-6.1 Sol (xhigh)52.9% ± 7.8\n5. ![](/labs/anthropic.svg)Claude Fable 5.1 (max)45.6% ± 5.3\n6. ![](/labs/openai.svg)GPT-6.1 Terra (xhigh)43.5% ± 6.8\n7. ![](/labs/google.svg)Gemini 4 Argon (high)41.3% ± 7.0\n8. ![](/labs/meta.svg)Muse Spark 1.3 (max)37.0% ± 4.6\n9. ![](/labs/moonshotai.svg)Kimi K3.1 Pro (max)34.6% ± 4.4\n10. ![](/labs/spacexai.svg)Grok 4.7 (xhigh)34.1% ± 4.4\n11. ![](/labs/alibaba.svg)Qwen 3.8 Max (xhigh)30.3% ± 4.2\n12. ![](/labs/thinkingmachines.svg)Inkling 2 Small (max)29.8% ± 3.7\n13. ![](/labs/zai.svg)GLM 5.3 (max)27.7% ± 3.8\n14. ![](/labs/deepseek.svg)DeepSeek V4.1 Flash (max)27.3% ± 3.9\n15. ![](/labs/openai.svg)GPT-6 Luna (xhigh)24.6% ± 4.5\n16. ![](/labs/mistral.svg)Mistral Large 4 (high)21.5% ± 4.4\n17. ![](/labs/deepseek.svg)DeepSeek V4 Pro (max)20.5% ± 4.1\n18. ![](/labs/nvidia.svg)Nemotron 3 Ultra (high)14.5% ± 3.3\n19. ![](/labs/thinkingmachines.svg)Inkling (max)13.1% ± 3.9\n20. ![](/labs/thinkingmachines.svg)Inkling Small (max)10.9% ± 3.8\n\n0%10%20%30%40%50%60%70%\n\nMean rubric score (%)\n\nSource: DeepFinance (Julius Research)·Updated Nov 5, 2026·[juliusresearch.ai](https://www.juliusresearch.ai)\n\nResults: mean rubric score (%) by model configuration with its 95% confidence interval, best first\n\n| Rank | Model | Lab | Mean rubric score with 95% confidence interval |\n| --- | --- | --- | --- |\n| 1 | Claude Opus 5.5 (max) | Anthropic | 54.8% mean rubric score, 95% CI 46.2 to 64.4 |\n| 2 | Claude Sonnet 5.5 (max) | Anthropic | 54.6% mean rubric score, 95% CI 46.0 to 63.9 |\n| 3 | GPT-6 Astra (xhigh) | OpenAI | 54.2% mean rubric score, 95% CI 46.4 to 62.6 |\n| 4 | GPT-6.1 Sol (xhigh) | OpenAI | 52.9% mean rubric score, 95% CI 45.4 to 61.0 |\n| 5 | Claude Fable 5.1 (max) | Anthropic | 45.6% mean rubric score, 95% CI 40.7 to 51.3 |\n| 6 | GPT-6.1 Terra (xhigh) | OpenAI | 43.5% mean rubric score, 95% CI 37.1 to 50.6 |\n| 7 | Gemini 4 Argon (high) | Google | 41.3% mean rubric score, 95% CI 34.7 to 48.8 |\n| 8 | Muse Spark 1.3 (max) | Meta | 37.0% mean rubric score, 95% CI 32.6 to 41.9 |\n| 9 | Kimi K3.1 Pro (max) | Moonshot AI | 34.6% mean rubric score, 95% CI 30.4 to 39.2 |\n| 10 | Grok 4.7 (xhigh) | SpaceXAI | 34.1% mean rubric score, 95% CI 29.9 to 38.7 |\n| 11 | Qwen 3.8 Max (xhigh) | Alibaba | 30.3% mean rubric score, 95% CI 26.3 to 34.6 |\n| 12 | Inkling 2 Small (max) | Thinking Machines | 29.8% mean rubric score, 95% CI 26.1 to 33.5 |\n| 13 | GLM 5.3 (max) | Z.ai | 27.7% mean rubric score, 95% CI 23.9 to 31.5 |\n| 14 | DeepSeek V4.1 Flash (max) | DeepSeek | 27.3% mean rubric score, 95% CI 23.6 to 31.3 |\n| 15 | GPT-6 Luna (xhigh) | OpenAI | 24.6% mean rubric score, 95% CI 20.1 to 29.2 |\n| 16 | Mistral Large 4 (high) | Mistral | 21.5% mean rubric score, 95% CI 17.2 to 25.9 |\n| 17 | DeepSeek V4 Pro (max) | DeepSeek | 20.5% mean rubric score, 95% CI 16.5 to 24.7 |\n| 18 | Nemotron 3 Ultra (high) | NVIDIA | 14.5% mean rubric score, 95% CI 11.3 to 17.8 |\n| 19 | Inkling (max) | Thinking Machines | 13.1% mean rubric score, 95% CI 9.3 to 17.2 |\n| 20 | Inkling Small (max) | Thinking Machines | 10.9% mean rubric score, 95% CI 7.3 to 14.8 |\n\nMean rubric score by model configuration on DeepFinance (21 tasks, 10 attempts per task). Rubric score is the percentage of rubric criteria satisfied. Error bars denote 95% bootstrap confidence intervals (100,000 resamples over tasks and attempts).\n\n### Score vs. cost\n\nMean rubric score against mean API cost per task. The dashed line marks the Pareto frontier: configurations not dominated on both score and cost.\n\nSortable DeepFinance results by model configuration: mean rubric score (higher is better), API cost per task and tokens per task (lower is better)\n\n| Rank | Model |  |  |  |\n| --- | --- | --- | --- | --- |\n| 1 | Claude Opus 5.5 max | 54.8 ± 9.1, 95% CI 46.2–64.4 | $8.63 | 10.5M |\n| 2 | Claude Sonnet 5.5 max | 54.6 ± 9.0, 95% CI 46.0–63.9 | $6.94 | 22.6M |\n| 3 | GPT-6 Astra xhigh | 54.2 ± 8.1, 95% CI 46.4–62.6 | $4.55 | 1.72M |\n| 4 | GPT-6.1 Sol xhigh | 52.9 ± 7.8, 95% CI 45.4–61.0 | $0.81 | 1.90M |\n| 5 | Claude Fable 5.1 max | 45.6 ± 5.3, 95% CI 40.7–51.3 | $12.02 | 3.15M |\n| 6 | GPT-6.1 Terra xhigh | 43.5 ± 6.8, 95% CI 37.1–50.6 | $0.38 | 2.15M |\n| 7 | Gemini 4 Argon high | 41.3 ± 7.0, 95% CI 34.7–48.8 | $2.85 | 6.69M |\n| 8 | Muse Spark 1.3 max | 37.0 ± 4.6, 95% CI 32.6–41.9 | $4.87 | 10.9M |\n| 9 | Kimi K3.1 Pro max | 34.6 ± 4.4, 95% CI 30.4–39.2 | $2.67 | 6.08M |\n| 10 | Grok 4.7 xhigh | 34.1 ± 4.4, 95% CI 29.9–38.7 | $4.71 | 5.90M |\n| 11 | Qwen 3.8 Max xhigh | 30.3 ± 4.2, 95% CI 26.3–34.6 | $4.99 | 12.6M |\n| 12 | Inkling 2 Small max | 29.8 ± 3.7, 95% CI 26.1–33.5 | $0.50 | 6.38M |\n| 13 | GLM 5.3 max | 27.7 ± 3.8, 95% CI 23.9–31.5 | $1.11 | 5.67M |\n| 14 | DeepSeek V4.1 Flash max | 27.3 ± 3.9, 95% CI 23.6–31.3 | $0.13 | 3.93M |\n| 15 | GPT-6 Luna xhigh | 24.6 ± 4.5, 95% CI 20.1–29.2 | $0.09 | 2.80M |\n| 16 | Mistral Large 4 high | 21.5 ± 4.4, 95% CI 17.2–25.9 | $3.69 | 5.04M |\n| 17 | DeepSeek V4 Pro max | 20.5 ± 4.1, 95% CI 16.5–24.7 | $1.50 | 2.64M |\n| 18 | Nemotron 3 Ultra high | 14.5 ± 3.3, 95% CI 11.3–17.8 | $0.64 | 2.83M |\n| 19 | Inkling max | 13.1 ± 3.9, 95% CI 9.3–17.2 | $0.64 | 2.40M |\n| 20 | Inkling Small max | 10.9 ± 3.8, 95% CI 7.3–14.8 | $0.15 | 1.09M |\n\n[Read the post: DeepFinance: Can AI diligence a $100MM deal?](/blog/deep-finance)\n","body_html":"<h2 id=\"deepfinance\">DeepFinance</h2>\n<p>An ultra long-horizon benchmark for measuring models on real-world Private Equity and Investment Banking workflows that would take a human professional between 20 and 30 hours to complete.</p>\n<p>tasks\n:   21</p>\n<p>average human time\n:   27.1 hours per task</p>\n<p>rubric criteria per task\n:   84</p>\n<p>files per data room\n:   19</p>\n<p>Mean rubric score (%)</p>\n<p>0%10%20%30%40%50%60%70%</p>\n<p>Julius Research</p>\n<ol><li><p>54.8</p><p> Claude Opus 5.5(max)</p></li><li><p>54.6</p><p> Claude Sonnet 5.5(max)</p></li><li><p>54.2</p><p> GPT-6 Astra(xhigh)</p></li><li><p>52.9</p><p> GPT-6.1 Sol(xhigh)</p></li><li><p>45.6</p><p> Claude Fable 5.1(max)</p></li><li><p>43.5</p><p> GPT-6.1 Terra(xhigh)</p></li><li><p>41.3</p><p> Gemini 4 Argon(high)</p></li><li><p>37.0</p><p> Muse Spark 1.3(max)</p></li><li><p>34.6</p><p> Kimi K3.1 Pro(max)</p></li><li><p>34.1</p><p>  Grok 4.7(xhigh)</p></li><li><p>30.3</p><p>  Qwen 3.8 Max(xhigh)</p></li><li><p>29.8</p><p>  Inkling 2 Small(max)</p></li><li><p>27.7</p><p>  GLM 5.3(max)</p></li><li><p>27.3</p><p>  DeepSeek V4.1 Flash(max)</p></li><li><p>24.6</p><p>  GPT-6 Luna(xhigh)</p></li><li><p>21.5</p><p>  Mistral Large 4(high)</p></li><li><p>20.5</p><p>  DeepSeek V4 Pro(max)</p></li><li><p>14.5</p><p>  Nemotron 3 Ultra(high)</p></li><li><p>13.1</p><p>  Inkling(max)</p></li><li><p>10.9</p><p>  Inkling Small(max)</p></li></ol>\n<p>Claude Opus 5.5· max</p>\n<p>54.8% ± 9.1</p>\n<p>Julius Research</p>\n<ol><li>Claude Opus 5.5 (max)54.8% ± 9.1</li><li>Claude Sonnet 5.5 (max)54.6% ± 9.0</li><li>GPT-6 Astra (xhigh)54.2% ± 8.1</li><li>GPT-6.1 Sol (xhigh)52.9% ± 7.8</li><li>Claude Fable 5.1 (max)45.6% ± 5.3</li><li>GPT-6.1 Terra (xhigh)43.5% ± 6.8</li><li>Gemini 4 Argon (high)41.3% ± 7.0</li><li>Muse Spark 1.3 (max)37.0% ± 4.6</li><li>Kimi K3.1 Pro (max)34.6% ± 4.4</li><li>Grok 4.7 (xhigh)34.1% ± 4.4</li><li>Qwen 3.8 Max (xhigh)30.3% ± 4.2</li><li>Inkling 2 Small (max)29.8% ± 3.7</li><li>GLM 5.3 (max)27.7% ± 3.8</li><li>DeepSeek V4.1 Flash (max)27.3% ± 3.9</li><li>GPT-6 Luna (xhigh)24.6% ± 4.5</li><li>Mistral Large 4 (high)21.5% ± 4.4</li><li>DeepSeek V4 Pro (max)20.5% ± 4.1</li><li>Nemotron 3 Ultra (high)14.5% ± 3.3</li><li>Inkling (max)13.1% ± 3.9</li><li>Inkling Small (max)10.9% ± 3.8</li></ol>\n<p>0%10%20%30%40%50%60%70%</p>\n<p>Mean rubric score (%)</p>\n<p>Source: DeepFinance (Julius Research)·Updated Nov 5, 2026·<a href=\"https://www.juliusresearch.ai\" rel=\"nofollow ugc noopener\">juliusresearch.ai</a></p>\n<p>Results: mean rubric score (%) by model configuration with its 95% confidence interval, best first</p>\n<div class=\"table-wrap\"><table><thead><tr><th>Rank</th><th>Model</th><th>Lab</th><th>Mean rubric score with 95% confidence interval</th></tr></thead><tbody><tr><td>1</td><td>Claude Opus 5.5 (max)</td><td>Anthropic</td><td>54.8% mean rubric score, 95% CI 46.2 to 64.4</td></tr><tr><td>2</td><td>Claude Sonnet 5.5 (max)</td><td>Anthropic</td><td>54.6% mean rubric score, 95% CI 46.0 to 63.9</td></tr><tr><td>3</td><td>GPT-6 Astra (xhigh)</td><td>OpenAI</td><td>54.2% mean rubric score, 95% CI 46.4 to 62.6</td></tr><tr><td>4</td><td>GPT-6.1 Sol (xhigh)</td><td>OpenAI</td><td>52.9% mean rubric score, 95% CI 45.4 to 61.0</td></tr><tr><td>5</td><td>Claude Fable 5.1 (max)</td><td>Anthropic</td><td>45.6% mean rubric score, 95% CI 40.7 to 51.3</td></tr><tr><td>6</td><td>GPT-6.1 Terra (xhigh)</td><td>OpenAI</td><td>43.5% mean rubric score, 95% CI 37.1 to 50.6</td></tr><tr><td>7</td><td>Gemini 4 Argon (high)</td><td>Google</td><td>41.3% mean rubric score, 95% CI 34.7 to 48.8</td></tr><tr><td>8</td><td>Muse Spark 1.3 (max)</td><td>Meta</td><td>37.0% mean rubric score, 95% CI 32.6 to 41.9</td></tr><tr><td>9</td><td>Kimi K3.1 Pro (max)</td><td>Moonshot AI</td><td>34.6% mean rubric score, 95% CI 30.4 to 39.2</td></tr><tr><td>10</td><td>Grok 4.7 (xhigh)</td><td>SpaceXAI</td><td>34.1% mean rubric score, 95% CI 29.9 to 38.7</td></tr><tr><td>11</td><td>Qwen 3.8 Max (xhigh)</td><td>Alibaba</td><td>30.3% mean rubric score, 95% CI 26.3 to 34.6</td></tr><tr><td>12</td><td>Inkling 2 Small (max)</td><td>Thinking Machines</td><td>29.8% mean rubric score, 95% CI 26.1 to 33.5</td></tr><tr><td>13</td><td>GLM 5.3 (max)</td><td>Z.ai</td><td>27.7% mean rubric score, 95% CI 23.9 to 31.5</td></tr><tr><td>14</td><td>DeepSeek V4.1 Flash (max)</td><td>DeepSeek</td><td>27.3% mean rubric score, 95% CI 23.6 to 31.3</td></tr><tr><td>15</td><td>GPT-6 Luna (xhigh)</td><td>OpenAI</td><td>24.6% mean rubric score, 95% CI 20.1 to 29.2</td></tr><tr><td>16</td><td>Mistral Large 4 (high)</td><td>Mistral</td><td>21.5% mean rubric score, 95% CI 17.2 to 25.9</td></tr><tr><td>17</td><td>DeepSeek V4 Pro (max)</td><td>DeepSeek</td><td>20.5% mean rubric score, 95% CI 16.5 to 24.7</td></tr><tr><td>18</td><td>Nemotron 3 Ultra (high)</td><td>NVIDIA</td><td>14.5% mean rubric score, 95% CI 11.3 to 17.8</td></tr><tr><td>19</td><td>Inkling (max)</td><td>Thinking Machines</td><td>13.1% mean rubric score, 95% CI 9.3 to 17.2</td></tr><tr><td>20</td><td>Inkling Small (max)</td><td>Thinking Machines</td><td>10.9% mean rubric score, 95% CI 7.3 to 14.8</td></tr></tbody></table></div>\n<p>Mean rubric score by model configuration on DeepFinance (21 tasks, 10 attempts per task). Rubric score is the percentage of rubric criteria satisfied. Error bars denote 95% bootstrap confidence intervals (100,000 resamples over tasks and attempts).</p>\n<h3 id=\"score-vs-cost\">Score vs. cost</h3>\n<p>Mean rubric score against mean API cost per task. The dashed line marks the Pareto frontier: configurations not dominated on both score and cost.</p>\n<p>Sortable DeepFinance results by model configuration: mean rubric score (higher is better), API cost per task and tokens per task (lower is better)</p>\n<div class=\"table-wrap\"><table><thead><tr><th>Rank</th><th>Model</th><th></th><th></th><th></th></tr></thead><tbody><tr><td>1</td><td>Claude Opus 5.5 max</td><td>54.8 ± 9.1, 95% CI 46.2–64.4</td><td>$8.63</td><td>10.5M</td></tr><tr><td>2</td><td>Claude Sonnet 5.5 max</td><td>54.6 ± 9.0, 95% CI 46.0–63.9</td><td>$6.94</td><td>22.6M</td></tr><tr><td>3</td><td>GPT-6 Astra xhigh</td><td>54.2 ± 8.1, 95% CI 46.4–62.6</td><td>$4.55</td><td>1.72M</td></tr><tr><td>4</td><td>GPT-6.1 Sol xhigh</td><td>52.9 ± 7.8, 95% CI 45.4–61.0</td><td>$0.81</td><td>1.90M</td></tr><tr><td>5</td><td>Claude Fable 5.1 max</td><td>45.6 ± 5.3, 95% CI 40.7–51.3</td><td>$12.02</td><td>3.15M</td></tr><tr><td>6</td><td>GPT-6.1 Terra xhigh</td><td>43.5 ± 6.8, 95% CI 37.1–50.6</td><td>$0.38</td><td>2.15M</td></tr><tr><td>7</td><td>Gemini 4 Argon high</td><td>41.3 ± 7.0, 95% CI 34.7–48.8</td><td>$2.85</td><td>6.69M</td></tr><tr><td>8</td><td>Muse Spark 1.3 max</td><td>37.0 ± 4.6, 95% CI 32.6–41.9</td><td>$4.87</td><td>10.9M</td></tr><tr><td>9</td><td>Kimi K3.1 Pro max</td><td>34.6 ± 4.4, 95% CI 30.4–39.2</td><td>$2.67</td><td>6.08M</td></tr><tr><td>10</td><td>Grok 4.7 xhigh</td><td>34.1 ± 4.4, 95% CI 29.9–38.7</td><td>$4.71</td><td>5.90M</td></tr><tr><td>11</td><td>Qwen 3.8 Max xhigh</td><td>30.3 ± 4.2, 95% CI 26.3–34.6</td><td>$4.99</td><td>12.6M</td></tr><tr><td>12</td><td>Inkling 2 Small max</td><td>29.8 ± 3.7, 95% CI 26.1–33.5</td><td>$0.50</td><td>6.38M</td></tr><tr><td>13</td><td>GLM 5.3 max</td><td>27.7 ± 3.8, 95% CI 23.9–31.5</td><td>$1.11</td><td>5.67M</td></tr><tr><td>14</td><td>DeepSeek V4.1 Flash max</td><td>27.3 ± 3.9, 95% CI 23.6–31.3</td><td>$0.13</td><td>3.93M</td></tr><tr><td>15</td><td>GPT-6 Luna xhigh</td><td>24.6 ± 4.5, 95% CI 20.1–29.2</td><td>$0.09</td><td>2.80M</td></tr><tr><td>16</td><td>Mistral Large 4 high</td><td>21.5 ± 4.4, 95% CI 17.2–25.9</td><td>$3.69</td><td>5.04M</td></tr><tr><td>17</td><td>DeepSeek V4 Pro max</td><td>20.5 ± 4.1, 95% CI 16.5–24.7</td><td>$1.50</td><td>2.64M</td></tr><tr><td>18</td><td>Nemotron 3 Ultra high</td><td>14.5 ± 3.3, 95% CI 11.3–17.8</td><td>$0.64</td><td>2.83M</td></tr><tr><td>19</td><td>Inkling max</td><td>13.1 ± 3.9, 95% CI 9.3–17.2</td><td>$0.64</td><td>2.40M</td></tr><tr><td>20</td><td>Inkling Small max</td><td>10.9 ± 3.8, 95% CI 7.3–14.8</td><td>$0.15</td><td>1.09M</td></tr></tbody></table></div>\n<p><a href=\"/blog/deep-finance\">Read the post: DeepFinance: Can AI diligence a $100MM deal?</a></p>","headings":[{"level":2,"text":"DeepFinance","id":"deepfinance"},{"level":3,"text":"Score vs. cost","id":"score-vs-cost"}]}}