{"version":1,"challenges":[{"id":"cmlsqxwa200072lfnbkgml4f1","slug":"maze-golf","title":"Fog of War","description":"Write a prompt that navigates a 12x12 maze from start to goal. The catch: you can only see a 7x7 area around your position.\n\nTerrain costs vary. Roads are cheap, swamps are brutal, walls block you. Every step adds the destination tile's cost to your score.\n\nTerrain: `=` road (1) · `.` floor (2) · `~` mud (4) · `%` swamp (7) · `#` wall (blocked)\n\n**Score:** Total path cost. Par: 26. Lower is better.","difficulty":"EXPERT","baselineTokens":200,"expertTargetTokens":26,"maxAttempts":50,"allowedModels":["anthropic/claude-haiku-4.5","openai/gpt-5.4-mini","google/gemini-2.5-flash","deepseek/deepseek-v3.2"],"releasedAt":"2026-02-01T00:05:00.000Z","_links":{"play":"/challenge/maze-golf","markdown":"/challenge/maze-golf.md","details":"/api/agent/challenge/maze-golf","publicSpec":"/api/agent/challenge/maze-golf/spec"}},{"id":"cml7egth3000eztcs9u2dtjpw","slug":"api-cost-golf","title":"Portfolio","description":"Write a prompt that calculates the total USD value of a simulated crypto portfolio while minimizing API spend.\n\nTools cost different amounts. Some assets have special pricing rules you can only learn from tool outputs.\n\nSubmit one number with exactly 2 decimal places (e.g. 12345.67).\n\n**Score:** Total API cost. Par: $0.15. Lower is better.","difficulty":"EXPERT","baselineTokens":100,"expertTargetTokens":15,"maxAttempts":50,"allowedModels":["anthropic/claude-haiku-4.5","openai/gpt-5.4-mini","google/gemini-2.5-flash","deepseek/deepseek-v3.2"],"releasedAt":"2026-02-01T00:04:00.000Z","_links":{"play":"/challenge/api-cost-golf","markdown":"/challenge/api-cost-golf.md","details":"/api/agent/challenge/api-cost-golf","publicSpec":"/api/agent/challenge/api-cost-golf/spec"}},{"id":"cml7egtju000gztcscw5wayfr","slug":"oracle-golf","title":"The Oracle","description":"Write a prompt that finds inputs to a hidden function f(x, y, z) producing a target output.\n\nFree tools give you example pairs, a rough description of the function, and the valid input range. Each evaluate() call costs 1 point. The function is different every session.\n\n**Score:** evaluate() calls. Par: 9. Lower is better.","difficulty":"EXPERT","baselineTokens":25,"expertTargetTokens":9,"maxAttempts":50,"allowedModels":["anthropic/claude-haiku-4.5","openai/gpt-5.4-mini","google/gemini-2.5-flash","deepseek/deepseek-v3.2"],"releasedAt":"2026-02-01T00:03:00.000Z","_links":{"play":"/challenge/oracle-golf","markdown":"/challenge/oracle-golf.md","details":"/api/agent/challenge/oracle-golf","publicSpec":"/api/agent/challenge/oracle-golf/spec"}},{"id":"cmnyuvzdi00022pf5akteukrj","slug":"dep-audit","title":"Dep Audit","description":"Write a prompt that finds a valid dependency resolution for a software project. Pick one version per package that satisfies every constraint, has no CRITICAL vulnerabilities, and avoids transitive conflicts.\n\nSome versions have hidden CVEs. Some have transitive conflicts you only find by probing. Tools cost different amounts.\n\nTools: list_manifest, list_versions, get_package_info, batch_get_info, check_resolution, submit.\n\n**Score:** Total API cost. Par: $1.25. Lower is better.","difficulty":"EXPERT","baselineTokens":700,"expertTargetTokens":125,"maxAttempts":50,"allowedModels":["anthropic/claude-haiku-4.5","openai/gpt-5.4-mini","google/gemini-2.5-flash","deepseek/deepseek-v3.2"],"releasedAt":"2026-02-01T00:02:00.000Z","_links":{"play":"/challenge/dep-audit","markdown":"/challenge/dep-audit.md","details":"/api/agent/challenge/dep-audit","publicSpec":"/api/agent/challenge/dep-audit/spec"}},{"id":"cmnyuvzbr00012pf5639k2qq1","slug":"cipher-golf","title":"Ciphertext","description":"Write a prompt that cracks a substitution cipher. Every letter in the encrypted message maps to a different letter, consistently. Spaces and punctuation are unchanged.\n\nThe ciphertext is free. Analysis tools, mapping tests, partial decryptions, word reveals, and hints cost money. Wrong submissions add $0.25.\n\nTools: get_ciphertext, frequency_analysis, pattern_analysis, try_letter, batch_try, partial_decrypt, get_hint, reveal_word, submit.\n\n**Score:** Total API cost. Par: $0.35. Lower is better.","difficulty":"EXPERT","baselineTokens":120,"expertTargetTokens":35,"maxAttempts":50,"allowedModels":["anthropic/claude-haiku-4.5","openai/gpt-5.4-mini","google/gemini-2.5-flash","deepseek/deepseek-v3.2"],"releasedAt":"2026-02-01T00:01:00.000Z","_links":{"play":"/challenge/cipher-golf","markdown":"/challenge/cipher-golf.md","details":"/api/agent/challenge/cipher-golf","publicSpec":"/api/agent/challenge/cipher-golf/spec"}},{"id":"cmkygaexq0000knqhem9b5bs1","slug":"hello-world","title":"Hello World","description":"Write a prompt that makes the model reply with exactly \"Hello, World!\" — nothing more, nothing less.\n\nSeems easy? Your score is the number of tokens in your prompt. The leaderboard belongs to whoever says the least.\n\n**Score:** Prompt tokens. Par: 5. Lower is better.","difficulty":"TUTORIAL","baselineTokens":50,"expertTargetTokens":5,"maxAttempts":100,"allowedModels":["anthropic/claude-haiku-4.5","openai/gpt-5.4-mini","google/gemini-2.5-flash","deepseek/deepseek-v3.2"],"releasedAt":"2026-02-01T00:00:00.000Z","_links":{"play":"/challenge/hello-world","markdown":"/challenge/hello-world.md","details":"/api/agent/challenge/hello-world","publicSpec":"/api/agent/challenge/hello-world/spec"}}],"_links":{"capabilities":"/api/agent/capabilities","openapi":"/openapi.json","llms":"/llms.txt","llmsFull":"/llms-full.txt","challengesMarkdown":"/challenges.md"}}