{"benchmark": {"slug": "hal-corebench-hard", "name": "CORE-Bench Hard (HAL)", "description": "CORE-Bench Hard on HAL measures whether agents can reproduce published computational results. Rows include accuracy and total API cost normalized to per-task cost in the catalog (`hal_site`).\n\nUseful for research-agent and notebook-style scaffolds (Claude Code, CORE-Agent, etc.) beyond pure issue-fixing.", "category": "eval", "split": null, "language": null, "task_count": 45, "homepage_url": "https://hal.cs.princeton.edu/corebench_hard", "paper_url": "https://arxiv.org/abs/2510.11977", "hf_dataset_id": null, "official_leaderboard_url": "https://hal.cs.princeton.edu/corebench_hard", "is_public_leaderboard": true, "harness_count": 3, "model_count": 24, "score_count": 47, "best_success_rate": 77.78}, "protocol": {"eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "temperature": null, "max_turns": null, "context_budget_tokens": null, "n": 47}, "rankings": {"benchmark": {"slug": "hal-corebench-hard", "name": "CORE-Bench Hard (HAL)", "description": "CORE-Bench Hard on HAL measures whether agents can reproduce published computational results. Rows include accuracy and total API cost normalized to per-task cost in the catalog (`hal_site`).\n\nUseful for research-agent and notebook-style scaffolds (Claude Code, CORE-Agent, etc.) beyond pure issue-fixing.", "category": "eval", "split": null, "language": null, "task_count": 45, "homepage_url": "https://hal.cs.princeton.edu/corebench_hard", "paper_url": "https://arxiv.org/abs/2510.11977", "hf_dataset_id": null, "official_leaderboard_url": "https://hal.cs.princeton.edu/corebench_hard", "is_public_leaderboard": true, "harness_count": null, "model_count": null, "score_count": null, "best_success_rate": null}, "sort": "success_rate", "rows": [{"rank": 1, "harness": {"slug": "claude-code", "name": "Claude Code", "vendor": "Anthropic", "scaffold_code": "cc", "repo_url": "https://github.com/anthropics/claude-code", "homepage_url": "https://github.com/anthropics/claude-code", "openrouter_icon_url": "https://claude.ai/favicon.ico", "leaderboard_icon_url": null}, "model": {"slug": "claude-4-5-opus", "name": "Claude 4.5 Opus", "vendor": "Anthropic"}, "success_rate": 77.78, "resolved_count": null, "total_count": null, "cost_usd": 1.9369, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 2, "harness": {"slug": "claude-code", "name": "Claude Code", "vendor": "Anthropic", "scaffold_code": "cc", "repo_url": "https://github.com/anthropics/claude-code", "homepage_url": "https://github.com/anthropics/claude-code", "openrouter_icon_url": "https://claude.ai/favicon.ico", "leaderboard_icon_url": null}, "model": {"slug": "claude-4-5-sonnet", "name": "Claude 4.5 Sonnet", "vendor": "Anthropic"}, "success_rate": 62.22, "resolved_count": null, "total_count": null, "cost_usd": 1.5184, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 3, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-4-1-opus", "name": "Claude 4.1 Opus", "vendor": "Anthropic"}, "success_rate": 51.11, "resolved_count": null, "total_count": null, "cost_usd": 9.1649, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 4, "harness": {"slug": "claude-code", "name": "Claude Code", "vendor": "Anthropic", "scaffold_code": "cc", "repo_url": "https://github.com/anthropics/claude-code", "homepage_url": "https://github.com/anthropics/claude-code", "openrouter_icon_url": "https://claude.ai/favicon.ico", "leaderboard_icon_url": null}, "model": {"slug": "claude-4-sonnet", "name": "Claude 4 Sonnet", "vendor": "Anthropic"}, "success_rate": 46.67, "resolved_count": null, "total_count": null, "cost_usd": 1.4573, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 5, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-4-5-sonnet-high", "name": "Claude 4.5 Sonnet (high)", "vendor": "Anthropic"}, "success_rate": 44.44, "resolved_count": null, "total_count": null, "cost_usd": 2.052, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 6, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-4-5-opus-high", "name": "Claude 4.5 Opus (high)", "vendor": "Anthropic"}, "success_rate": 42.22, "resolved_count": null, "total_count": null, "cost_usd": 3.3924, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 7, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-4-5-opus", "name": "Claude 4.5 Opus", "vendor": "Anthropic"}, "success_rate": 42.22, "resolved_count": null, "total_count": null, "cost_usd": 3.7553, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 8, "harness": {"slug": "claude-code", "name": "Claude Code", "vendor": "Anthropic", "scaffold_code": "cc", "repo_url": "https://github.com/anthropics/claude-code", "homepage_url": "https://github.com/anthropics/claude-code", "openrouter_icon_url": "https://claude.ai/favicon.ico", "leaderboard_icon_url": null}, "model": {"slug": "claude-4-1-opus", "name": "Claude 4.1 Opus", "vendor": "Anthropic"}, "success_rate": 42.22, "resolved_count": null, "total_count": null, "cost_usd": 7.3731, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 9, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-4-1-opus-high", "name": "Claude 4.1 Opus (high)", "vendor": "Anthropic"}, "success_rate": 42.22, "resolved_count": null, "total_count": null, "cost_usd": 11.3322, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 10, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "gemini-3-pro-preview", "name": "Gemini 3 Pro Preview", "vendor": "Google"}, "success_rate": 40.0, "resolved_count": null, "total_count": null, "cost_usd": 1.9244, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 11, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-3-7-sonnet-high", "name": "Claude 3.7 Sonnet (high)", "vendor": "Anthropic"}, "success_rate": 37.78, "resolved_count": null, "total_count": null, "cost_usd": 1.47, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 12, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-4-5-sonnet", "name": "Claude 4.5 Sonnet", "vendor": "Anthropic"}, "success_rate": 37.78, "resolved_count": null, "total_count": null, "cost_usd": 2.1589, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 13, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-3-7-sonnet", "name": "Claude 3.7 Sonnet", "vendor": "Anthropic"}, "success_rate": 35.56, "resolved_count": null, "total_count": null, "cost_usd": 1.6231, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 14, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "gemini-3-pro-preview", "name": "Gemini 3 Pro Preview", "vendor": "Google"}, "success_rate": 35.56, "resolved_count": null, "total_count": null, "cost_usd": 2.2504, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 15, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-4-1-opus", "name": "Claude 4.1 Opus", "vendor": "Anthropic"}, "success_rate": 35.56, "resolved_count": null, "total_count": null, "cost_usd": 8.3358, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 16, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-4-5-sonnet", "name": "Claude 4.5 Sonnet", "vendor": "Anthropic"}, "success_rate": 33.33, "resolved_count": null, "total_count": null, "cost_usd": 1.8931, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 17, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-4-sonnet-high", "name": "Claude 4 Sonnet (high)", "vendor": "Anthropic"}, "success_rate": 33.33, "resolved_count": null, "total_count": null, "cost_usd": 2.2329, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 18, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "gpt-4-1", "name": "GPT-4.1", "vendor": "openai"}, "success_rate": 33.33, "resolved_count": null, "total_count": null, "cost_usd": 2.3858, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 19, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-4-5-opus", "name": "Claude 4.5 Opus", "vendor": "Anthropic"}, "success_rate": 33.33, "resolved_count": null, "total_count": null, "cost_usd": 2.8313, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 20, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-4-1-opus-high", "name": "Claude 4.1 Opus (high)", "vendor": "Anthropic"}, "success_rate": 33.33, "resolved_count": null, "total_count": null, "cost_usd": 7.966, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 21, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-3-7-sonnet", "name": "Claude 3.7 Sonnet", "vendor": "Anthropic"}, "success_rate": 31.11, "resolved_count": null, "total_count": null, "cost_usd": 1.2587, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 22, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-4-5-opus-high", "name": "Claude 4.5 Opus (high)", "vendor": "Anthropic"}, "success_rate": 31.11, "resolved_count": null, "total_count": null, "cost_usd": 2.4973, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 23, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-4-sonnet", "name": "Claude 4 Sonnet", "vendor": "Anthropic"}, "success_rate": 28.89, "resolved_count": null, "total_count": null, "cost_usd": 1.1171, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 24, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-4-5-sonnet-high", "name": "Claude 4.5 Sonnet (high)", "vendor": "Anthropic"}, "success_rate": 28.89, "resolved_count": null, "total_count": null, "cost_usd": 1.9504, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 25, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "gpt-5-medium", "name": "GPT-5 (medium)", "vendor": "OpenAI"}, "success_rate": 26.67, "resolved_count": null, "total_count": null, "cost_usd": 0.7058, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 26, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-3-7-sonnet-high", "name": "Claude 3.7 Sonnet (high)", "vendor": "Anthropic"}, "success_rate": 24.44, "resolved_count": null, "total_count": null, "cost_usd": 1.6104, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 27, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "o3", "name": "o3", "vendor": "openai"}, "success_rate": 24.44, "resolved_count": null, "total_count": null, "cost_usd": 2.6771, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 28, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "gpt-4-1", "name": "GPT-4.1", "vendor": "openai"}, "success_rate": 22.22, "resolved_count": null, "total_count": null, "cost_usd": 1.296, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 29, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "o3", "name": "o3", "vendor": "openai"}, "success_rate": 22.22, "resolved_count": null, "total_count": null, "cost_usd": 1.9631, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 30, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "gemini-2-5-pro-preview", "name": "Gemini 2.5 Pro Preview", "vendor": "google"}, "success_rate": 22.22, "resolved_count": null, "total_count": null, "cost_usd": 4.052, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 31, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "deepseek-v3-1-august-2025", "name": "DeepSeek V3.1 (August 2025)", "vendor": "DeepSeek"}, "success_rate": 20.0, "resolved_count": null, "total_count": null, "cost_usd": 0.2789, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 32, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "o4-mini", "name": "o4 mini", "vendor": "openai"}, "success_rate": 17.78, "resolved_count": null, "total_count": null, "cost_usd": 0.7064, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 33, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "deepseek-v3-march-2025", "name": "DeepSeek V3 (March 2025)", "vendor": "DeepSeek"}, "success_rate": 17.78, "resolved_count": null, "total_count": null, "cost_usd": 0.5613, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 34, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "o4-mini", "name": "o4 mini", "vendor": "openai"}, "success_rate": 15.56, "resolved_count": null, "total_count": null, "cost_usd": 0.5, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 35, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "gpt-oss-120b-august-2025", "name": "GPT OSS 120B (August 2025)", "vendor": "OpenAI"}, "success_rate": 11.11, "resolved_count": null, "total_count": null, "cost_usd": 0.0936, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 36, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "gpt-oss-120b-high-august-2025", "name": "GPT OSS 120B High (August 2025)", "vendor": "OpenAI"}, "success_rate": 11.11, "resolved_count": null, "total_count": null, "cost_usd": 0.0936, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 37, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "gemini-2-0-flash", "name": "Gemini 2.0 Flash", "vendor": "Google"}, "success_rate": 11.11, "resolved_count": null, "total_count": null, "cost_usd": 0.2769, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 38, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "gpt-5-medium", "name": "GPT-5 (medium)", "vendor": "OpenAI"}, "success_rate": 11.11, "resolved_count": null, "total_count": null, "cost_usd": 0.6611, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 39, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "claude-4-5-haiku", "name": "Claude 4.5 Haiku", "vendor": "Anthropic"}, "success_rate": 11.11, "resolved_count": null, "total_count": null, "cost_usd": 0.9762, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 40, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "gpt-oss-120b-high-august-2025", "name": "GPT OSS 120B High (August 2025)", "vendor": "OpenAI"}, "success_rate": 8.89, "resolved_count": null, "total_count": null, "cost_usd": 0.0456, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 41, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "gpt-oss-120b-august-2025", "name": "GPT OSS 120B (August 2025)", "vendor": "OpenAI"}, "success_rate": 8.89, "resolved_count": null, "total_count": null, "cost_usd": 0.062, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 42, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "deepseek-v3-march-2025", "name": "DeepSeek V3 (March 2025)", "vendor": "DeepSeek"}, "success_rate": 8.89, "resolved_count": null, "total_count": null, "cost_usd": 0.1042, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 43, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "deepseek-r1-may-2025", "name": "DeepSeek R1 (May 2025)", "vendor": "DeepSeek"}, "success_rate": 8.89, "resolved_count": null, "total_count": null, "cost_usd": 0.1727, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 44, "harness": {"slug": "core-agent", "name": "CORE-Agent", "vendor": null, "scaffold_code": null, "repo_url": null, "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "deepseek-r1-january-2025", "name": "DeepSeek R1 (January 2025)", "vendor": "DeepSeek"}, "success_rate": 6.67, "resolved_count": null, "total_count": null, "cost_usd": 1.8024, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 45, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "deepseek-r1-january-2025", "name": "DeepSeek R1 (January 2025)", "vendor": "DeepSeek"}, "success_rate": 4.45, "resolved_count": null, "total_count": null, "cost_usd": 0.5544, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 46, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "gemini-2-0-flash", "name": "Gemini 2.0 Flash", "vendor": "Google"}, "success_rate": 4.44, "resolved_count": null, "total_count": null, "cost_usd": 0.1569, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}, {"rank": 47, "harness": {"slug": "hal-generalist-agent", "name": "HAL Generalist Agent", "vendor": "Princeton PLI", "scaffold_code": null, "repo_url": "https://github.com/princeton-pli/hal-harness", "homepage_url": "https://hal.cs.princeton.edu/", "openrouter_icon_url": null, "leaderboard_icon_url": null}, "model": {"slug": "gemini-2-5-pro-preview", "name": "Gemini 2.5 Pro Preview", "vendor": "google"}, "success_rate": 4.44, "resolved_count": null, "total_count": null, "cost_usd": 0.6751, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "is_harnessrl_measured": false, "eval_protocol": "HAL CORE-Bench Hard (HAL) accuracy", "source": "hal_site", "source_url": "https://hal.cs.princeton.edu/corebench_hard", "observed_at": null}]}}