{"harness": {"slug": "codestory-midwit-agent", "name": "Codestory Midwit Agent", "vendor": "CodeStory", "description": "CodeStory's Midwit agent appears on SWE-bench Verified as a community submission harness. CodeStory builds IDE-adjacent coding agents; public leaderboard rows reflect a specific eval configuration and model pairing at submission time.\n\n`swe_bench_site` ingest. Check `source_url` for swebench.com metadata and the model column for the foundation model used.", "card_summary": "CodeStory Midwit agent \u2014 community Verified submission harness.", "homepage_url": "https://codestory.ai", "repo_url": null, "docs_url": null, "scaffold_code": null, "harbor_agent_name": null, "api_protocol": null, "status": "validated", "supports_rl": true, "score_count": 2, "best_success_rate": 62.2, "github_stars": null, "popularity_tier": null, "hrl_score": null, "hrl_rank": null, "catalog_token_volume": null, "openrouter_icon_url": null, "leaderboard_icon_url": "https://www.swebench.com/img/logos/20240702_codestory_aide_mixed.png", "aa_coding_index": null, "aa_mean_cost_usd": null, "aa_mean_total_tokens": null}, "scores": [{"success_rate": 62.2, "resolved_count": 311, "total_count": 500, "cost_usd": null, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "harness_version": null, "trial_count": null, "is_harnessrl_measured": false, "eval_protocol": "SWE-bench Verified public submission", "source": "swe_bench_site", "source_url": "https://aide.dev/", "observed_at": "2024-12-21T00:00:00+00:00", "harness": {"slug": "codestory-midwit-agent", "name": "Codestory Midwit Agent", "vendor": "CodeStory", "repo_url": null, "openrouter_icon_url": null}, "model": {"slug": "swe-search", "name": "SWE Search", "vendor": "Princeton NLP"}, "benchmark": {"slug": "swe-bench-verified", "name": "SWE-bench Verified"}}, {"success_rate": 43.0, "resolved_count": 129, "total_count": 300, "cost_usd": null, "input_tokens": null, "output_tokens": null, "latency_ms": null, "temperature": null, "max_turns": null, "context_budget_tokens": null, "harness_version": null, "trial_count": null, "is_harnessrl_measured": false, "eval_protocol": "SWE-bench Lite public submission", "source": "swe_bench_site", "source_url": "https://aide.dev/", "observed_at": "2024-07-02T00:00:00+00:00", "harness": {"slug": "codestory-midwit-agent", "name": "Codestory Midwit Agent", "vendor": "CodeStory", "repo_url": null, "openrouter_icon_url": null}, "model": {"slug": "mixed-models", "name": "mixed-models", "vendor": "Mixed"}, "benchmark": {"slug": "swe-bench-lite", "name": "SWE-bench Lite"}}], "benchmarks": [{"slug": "swe-bench-lite", "name": "SWE-bench Lite", "category": "eval"}, {"slug": "swe-bench-verified", "name": "SWE-bench Verified", "category": "eval"}]}