Tests private equity and investment banking work such as valuation, due diligence and financial modeling, helping you assess AI support for demanding client analysis. Scores reflect deliverables, workflow and judgment in model-specific agent setups.
| Model / creator | Version | Reported score |
|---|---|---|
 Claude Opus 5.5Anthropic · Proprietary | November 5, 2026 / native agent harnesses | 54.8%95% CI 46.2%–64.4% |
| Claude Sonnet 5.5Anthropic · Proprietary | November 5, 2026 / native agent harnesses | 54.6%95% CI 46%–63.9% |
 GPT-6 AstraOpenAI · Proprietary | November 5, 2026 / native agent harnesses | 54.2%95% CI 46.4%–62.6% |
| GPT-6.1 SolOpenAI · Proprietary | November 5, 2026 / native agent harnesses | 52.9%95% CI 45.4%–61% |
| Claude Fable 5.1Anthropic · Proprietary | November 5, 2026 / native agent harnesses | 45.6%95% CI 40.7%–51.3% |
| GPT-6.1 TerraOpenAI · Proprietary | November 5, 2026 / native agent harnesses | 43.5%95% CI 37.1%–50.6% |
 Gemini 4 ArgonGoogle · Proprietary | November 5, 2026 / native agent harnesses | 41.3%95% CI 34.7%–48.8% |
 Muse Spark 1.3Meta · Proprietary | November 5, 2026 / native agent harnesses | 37%95% CI 32.6%–41.9% |
 Kimi K3.1 ProMoonshot AI · Open source / open weights | November 5, 2026 / native agent harnesses | 34.6%95% CI 30.4%–39.2% |
 Grok 4.7xAI · Proprietary | November 5, 2026 / native agent harnesses | 34.1%95% CI 29.9%–38.7% |
 Qwen 3.8 MaxQwen · Not classified | November 5, 2026 / native agent harnesses | 30.3%95% CI 26.3%–34.6% |
 GLM-5.3Z.ai · Open source / open weights | November 5, 2026 / native agent harnesses | 27.7%95% CI 23.9%–31.5% |
 DeepSeek V4.1 FlashDeepSeek · Open source / open weights | November 5, 2026 / native agent harnesses | 27.3%95% CI 23.6%–31.3% |
| GPT-6 LunaOpenAI · Proprietary | November 5, 2026 / native agent harnesses | 24.6%95% CI 20.1%–29.2% |
 Mistral Large 4Mistral · Open source / open weights | November 5, 2026 / native agent harnesses | 21.5%95% CI 17.2%–25.9% |
| DeepSeek V4 ProDeepSeek · Open source / open weights | November 5, 2026 / native agent harnesses | 20.5%95% CI 16.5%–24.7% |
attempts: 10 · metric: Mean rubric score · native harness: claude-code 2.1.318 · reasoning effort: max · Julius Research native-agent protocol; claude-code 2.1.318; effort=max; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗attempts: 10 · metric: Mean rubric score · native harness: claude-code 2.1.318 · reasoning effort: max · Julius Research native-agent protocol; claude-code 2.1.318; effort=max; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗attempts: 10 · metric: Mean rubric score · native harness: codex 0.166.0 · reasoning effort: xhigh · Julius Research native-agent protocol; codex 0.166.0; effort=xhigh; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗attempts: 10 · metric: Mean rubric score · native harness: codex 0.166.0 · reasoning effort: xhigh · Julius Research native-agent protocol; codex 0.166.0; effort=xhigh; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗attempts: 10 · metric: Mean rubric score · native harness: claude-code 2.1.318 · reasoning effort: max · Julius Research native-agent protocol; claude-code 2.1.318; effort=max; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗attempts: 10 · metric: Mean rubric score · native harness: codex 0.166.0 · reasoning effort: xhigh · Julius Research native-agent protocol; codex 0.166.0; effort=xhigh; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗attempts: 10 · metric: Mean rubric score · native harness: opencode-ai 1.19.3 · reasoning effort: high · Julius Research native-agent protocol; opencode-ai 1.19.3; effort=high; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗attempts: 10 · metric: Mean rubric score · native harness: opencode-ai 1.19.3 · reasoning effort: max · Julius Research native-agent protocol; opencode-ai 1.19.3; effort=max; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗attempts: 10 · metric: Mean rubric score · native harness: kimi-code 2.4.1 · reasoning effort: max · Julius Research native-agent protocol; kimi-code 2.4.1; effort=max; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗attempts: 10 · metric: Mean rubric score · native harness: x.ai 1.0.70 · reasoning effort: xhigh · Julius Research native-agent protocol; x.ai 1.0.70; effort=xhigh; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗attempts: 10 · metric: Mean rubric score · native harness: qwen-code 0.26.6 · reasoning effort: xhigh · Julius Research native-agent protocol; qwen-code 0.26.6; effort=xhigh; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗attempts: 10 · metric: Mean rubric score · native harness: opencode-ai 1.19.3 · reasoning effort: max · Julius Research native-agent protocol; opencode-ai 1.19.3; effort=max; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗attempts: 10 · metric: Mean rubric score · native harness: opencode-ai 1.19.3 · reasoning effort: max · Julius Research native-agent protocol; opencode-ai 1.19.3; effort=max; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗attempts: 10 · metric: Mean rubric score · native harness: codex 0.166.0 · reasoning effort: xhigh · Julius Research native-agent protocol; codex 0.166.0; effort=xhigh; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗attempts: 10 · metric: Mean rubric score · native harness: mistral-vibe 2.24.3 · reasoning effort: high · Julius Research native-agent protocol; mistral-vibe 2.24.3; effort=high; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗attempts: 10 · metric: Mean rubric score · native harness: opencode-ai 1.19.3 · reasoning effort: max · Julius Research native-agent protocol; opencode-ai 1.19.3; effort=max; 21 tasks, 10 successful attempts per task; native tools and web search enabled, ask-user disabled. Harnesses vary by model. · tasks: 21 · Tools enabled
Julius Research ↗Tests rebuilding parts, assemblies and circuit boards from drawings and images, helping you assess AI support for client design and prototyping work.
AutomationBench-AATests how much of a business workflow an agent can complete without breaking its rules, helping you identify opportunities to reduce manual work across apps.
FrontierCodeTests whether an agent can produce software changes that meet functional and review requirements, helping you assess how much implementation work you could delegate before human code review.
