{"benchmark": {"slug": "hal-reliability-gaia", "name": "HAL Reliability \u2014 GAIA", "description": "HAL's reliability dashboard evaluates agents on GAIA with repeated runs and multi-dimensional metrics beyond single-shot accuracy. Catalog stores the mean of HAL's four dimension aggregates as `success_rate` (`hal_reliability_site`, harness `hal-generalist-agent`).\n\nSee hal.cs.princeton.edu/reliability/ for full consistency, predictability, robustness, and safety breakdowns.", "category": "eval", "split": null, "language": null, "task_count": null, "homepage_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "paper_url": "https://arxiv.org/abs/2510.11977", "hf_dataset_id": null, "official_leaderboard_url": "https://hal.cs.princeton.edu/reliability/", "is_public_leaderboard": true, "harness_count": 1, "model_count": 15, "score_count": 15, "best_success_rate": 96.25}, "protocol": {"eval_protocol": "HAL reliability composite (mean of 4 dimension aggregates)", "temperature": null, "max_turns": null, "context_budget_tokens": null, "n": 15}, "rankings": {"benchmark": {"slug": "hal-reliability-gaia", "name": "HAL Reliability \u2014 GAIA", "description": "HAL's reliability dashboard evaluates agents on GAIA with repeated runs and multi-dimensional metrics beyond single-shot accuracy. Catalog stores the mean of HAL's four dimension aggregates as `success_rate` (`hal_reliability_site`, harness `hal-generalist-agent`).\n\nSee hal.cs.princeton.edu/reliability/ for full consistency, predictability, robustness, and safety breakdowns.", "category": "eval", "split": null, "language": null, "task_count": null, "homepage_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "paper_url": "https://arxiv.org/abs/2510.11977", "hf_dataset_id": null, "official_leaderboard_url": "https://hal.cs.princeton.edu/reliability/", "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": "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": 96.25, "resolved_count": null, "total_count": null, "cost_usd": null, "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 reliability composite (mean of 4 dimension aggregates)", "source": "hal_reliability_site", "source_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "observed_at": null}, {"rank": 2, "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-sonnet", "name": "Claude 4 Sonnet", "vendor": "Anthropic"}, "success_rate": 91.0, "resolved_count": null, "total_count": null, "cost_usd": null, "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 reliability composite (mean of 4 dimension aggregates)", "source": "hal_reliability_site", "source_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "observed_at": null}, {"rank": 3, "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-4-7-opus", "name": "Claude 4.7 Opus", "vendor": "Anthropic"}, "success_rate": 90.75, "resolved_count": null, "total_count": null, "cost_usd": null, "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 reliability composite (mean of 4 dimension aggregates)", "source": "hal_reliability_site", "source_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "observed_at": null}, {"rank": 4, "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-3-5-flash", "name": "Gemini 3.5 Flash", "vendor": "google"}, "success_rate": 90.25, "resolved_count": null, "total_count": null, "cost_usd": null, "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 reliability composite (mean of 4 dimension aggregates)", "source": "hal_reliability_site", "source_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "observed_at": null}, {"rank": 5, "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-3-1-pro", "name": "Gemini 3.1 Pro", "vendor": "Google"}, "success_rate": 89.75, "resolved_count": null, "total_count": null, "cost_usd": null, "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 reliability composite (mean of 4 dimension aggregates)", "source": "hal_reliability_site", "source_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "observed_at": null}, {"rank": 6, "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-5-5", "name": "GPT-5.5", "vendor": "openai"}, "success_rate": 89.75, "resolved_count": null, "total_count": null, "cost_usd": null, "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 reliability composite (mean of 4 dimension aggregates)", "source": "hal_reliability_site", "source_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "observed_at": null}, {"rank": 7, "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-3-5-haiku", "name": "Claude 3.5 Haiku", "vendor": "Anthropic"}, "success_rate": 89.0, "resolved_count": null, "total_count": null, "cost_usd": null, "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 reliability composite (mean of 4 dimension aggregates)", "source": "hal_reliability_site", "source_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "observed_at": null}, {"rank": 8, "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", "name": "Gemini 2.5 Pro", "vendor": "google"}, "success_rate": 88.25, "resolved_count": null, "total_count": null, "cost_usd": null, "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 reliability composite (mean of 4 dimension aggregates)", "source": "hal_reliability_site", "source_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "observed_at": null}, {"rank": 9, "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": "o1", "name": "o1", "vendor": "openai"}, "success_rate": 87.75, "resolved_count": null, "total_count": null, "cost_usd": null, "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 reliability composite (mean of 4 dimension aggregates)", "source": "hal_reliability_site", "source_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "observed_at": null}, {"rank": 10, "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", "name": "GPT-4", "vendor": "openai"}, "success_rate": 87.5, "resolved_count": null, "total_count": null, "cost_usd": null, "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 reliability composite (mean of 4 dimension aggregates)", "source": "hal_reliability_site", "source_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "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": "gemini-2-5-flash", "name": "Gemini 2.5 Flash", "vendor": "google"}, "success_rate": 85.5, "resolved_count": null, "total_count": null, "cost_usd": null, "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 reliability composite (mean of 4 dimension aggregates)", "source": "hal_reliability_site", "source_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "observed_at": null}, {"rank": 12, "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-4o-mini", "name": "GPT-4o mini", "vendor": "openai"}, "success_rate": 84.75, "resolved_count": null, "total_count": null, "cost_usd": null, "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 reliability composite (mean of 4 dimension aggregates)", "source": "hal_reliability_site", "source_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "observed_at": null}, {"rank": 13, "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-5-2-medium", "name": "GPT-5.2 (medium)", "vendor": "OpenAI"}, "success_rate": 83.5, "resolved_count": null, "total_count": null, "cost_usd": null, "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 reliability composite (mean of 4 dimension aggregates)", "source": "hal_reliability_site", "source_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "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": "gpt-5-2", "name": "GPT-5.2", "vendor": "openai"}, "success_rate": 80.0, "resolved_count": null, "total_count": null, "cost_usd": null, "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 reliability composite (mean of 4 dimension aggregates)", "source": "hal_reliability_site", "source_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "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-3-haiku", "name": "Claude 3 Haiku", "vendor": "anthropic"}, "success_rate": 78.75, "resolved_count": null, "total_count": null, "cost_usd": null, "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 reliability composite (mean of 4 dimension aggregates)", "source": "hal_reliability_site", "source_url": "https://hal.cs.princeton.edu/reliability/benchmark/gaia/", "observed_at": null}]}}