{"cards":[{"repository":{"id":1298096233,"fullName":"mupt-ai/self-bench","description":"Benchmark coding agents and models on your own repository with private Harbor-formatted evals built from its merged pull requests.","stars":31,"ownerAvatarUrl":"https://avatars.githubusercontent.com/u/154857038?v=4"},"publisher":{"kind":"org","login":"mupt-ai","avatarUrl":"https://avatars.githubusercontent.com/u/154857038?v=4"},"releaseId":"502e232b-214d-4d5e-98eb-b6d40dfd8a1d","releasedAt":"2026-10-17T18:58:14.939Z","tasks":38,"settings":7,"picks":[{"setting":{"id":"gpt-6-luna|codex|high","model":{"label":"GPT-6 Luna"},"accuracy":47.368421052631575,"costPerTaskUsd":0.027170495657894738,"costedTasks":38},"roles":["cheapest"]},{"setting":{"id":"gpt-6.1-sol|codex|high","model":{"label":"GPT-6.1 Sol"},"accuracy":76.31578947368422,"costPerTaskUsd":0.4287800578947369,"costedTasks":38},"roles":["mostAccurate"]}],"frontier":[{"id":"gpt-6-luna|codex|high","model":{"label":"GPT-6 Luna"},"accuracy":47.368421052631575,"costPerTaskUsd":0.027170495657894738,"costedTasks":38},{"id":"deepseek/deepseek-v4.1-flash-1017|pi|high","model":{"label":"DeepSeek V4.1 Flash 1017"},"accuracy":71.05263157894737,"costPerTaskUsd":0.13343346710526313,"costedTasks":38},{"id":"gpt-6.1-sol|codex|high","model":{"label":"GPT-6.1 Sol"},"accuracy":76.31578947368422,"costPerTaskUsd":0.4287800578947369,"costedTasks":38}],"others":[{"id":"claude-opus-5-5|claude-code|high","model":{"label":"Claude Opus 5.5"},"accuracy":71.05263157894737,"costPerTaskUsd":1.5580841157894736,"costedTasks":38},{"id":"gpt-6.1-sol|codex|low","model":{"label":"GPT-6.1 Sol"},"accuracy":65.78947368421053,"costPerTaskUsd":0.21174511052631578,"costedTasks":38},{"id":"kimi-k3|codex|high","model":{"label":"Kimi K3"},"accuracy":63.1578947368421,"costPerTaskUsd":1.546290947368421,"costedTasks":38},{"id":"glm-5.3|codex|high","model":{"label":"GLM 5.3"},"accuracy":52.63157894736842,"costPerTaskUsd":1.3325128121052634,"costedTasks":38}],"endorsed":false,"defaultLine":true},{"repository":{"id":70107786,"fullName":"vercel/next.js","description":"The React Framework","stars":143053,"ownerAvatarUrl":"https://avatars.githubusercontent.com/u/14985020?v=4"},"publisher":{"kind":"org","login":"mupt-ai","avatarUrl":"https://avatars.githubusercontent.com/u/154857038?v=4"},"releaseId":"661c6566-8013-4a50-9bbc-48630ccbf47e","releasedAt":"2026-10-17T18:30:53.192Z","tasks":38,"settings":13,"picks":[{"setting":{"id":"gpt-6-luna|codex|high","model":{"label":"GPT-6 Luna"},"accuracy":39.473684210526315,"costPerTaskUsd":0.04176399052631579,"costedTasks":38},"roles":["cheapest"]},{"setting":{"id":"gpt-6.1-sol|codex|high","model":{"label":"GPT-6.1 Sol"},"accuracy":84.21052631578947,"costPerTaskUsd":0.9094339499999998,"costedTasks":38},"roles":["mostAccurate"]}],"frontier":[{"id":"gpt-6-luna|codex|high","model":{"label":"GPT-6 Luna"},"accuracy":39.473684210526315,"costPerTaskUsd":0.04176399052631579,"costedTasks":38},{"id":"claude-haiku-5-5|claude-code|low","model":{"label":"Claude Haiku 5.5"},"accuracy":50,"costPerTaskUsd":0.07115440789473682,"costedTasks":38},{"id":"deepseek/deepseek-v4.1-flash-1017|pi|high","model":{"label":"DeepSeek V4.1 Flash 1017"},"accuracy":71.05263157894737,"costPerTaskUsd":0.10139138936842104,"costedTasks":38},{"id":"gpt-6.1-sol|codex|low","model":{"label":"GPT-6.1 Sol"},"accuracy":78.94736842105263,"costPerTaskUsd":0.5620314052631579,"costedTasks":38},{"id":"gpt-6.1-sol|codex|high","model":{"label":"GPT-6.1 Sol"},"accuracy":84.21052631578947,"costPerTaskUsd":0.9094339499999998,"costedTasks":38}],"others":[{"id":"claude-opus-5-5|claude-code|low","model":{"label":"Claude Opus 5.5"},"accuracy":68.42105263157895,"costPerTaskUsd":0.7749697842105266,"costedTasks":38},{"id":"gpt-6-sol|codex|high","model":{"label":"GPT-6 Sol"},"accuracy":65.78947368421053,"costPerTaskUsd":2.0650251921052627,"costedTasks":38},{"id":"claude-haiku-5-5|claude-code|high","model":{"label":"Claude Haiku 5.5"},"accuracy":60.526315789473685,"costPerTaskUsd":0.6261940331578948,"costedTasks":38},{"id":"gpt-6-sol|codex|low","model":{"label":"GPT-6 Sol"},"accuracy":60.526315789473685,"costPerTaskUsd":0.8957784342105263,"costedTasks":38},{"id":"glm-5.3|pi|high","model":{"label":"GLM 5.3"},"accuracy":57.89473684210527,"costPerTaskUsd":5.255174198421052,"costedTasks":38}],"endorsed":false,"defaultLine":true},{"repository":{"id":235901813,"fullName":"PostHog/posthog","description":":hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.","stars":40321,"ownerAvatarUrl":"https://avatars.githubusercontent.com/u/60330232?v=4"},"publisher":{"kind":"org","login":"mupt-ai","avatarUrl":"https://avatars.githubusercontent.com/u/154857038?v=4"},"releaseId":"d16a5176-9a73-454a-9afa-fa62290c4eae","releasedAt":"2026-10-17T18:29:43.764Z","tasks":35,"settings":14,"picks":[{"setting":{"id":"gpt-6-luna|codex|low","model":{"label":"GPT-6 Luna"},"accuracy":25.71428571428571,"costPerTaskUsd":0.03371057214285714,"costedTasks":35},"roles":["cheapest"]},{"setting":{"id":"claude-opus-5-5|claude-code|high","model":{"label":"Claude Opus 5.5"},"accuracy":68.57142857142857,"costPerTaskUsd":2.3308018285714294,"costedTasks":35},"roles":["mostAccurate"]}],"frontier":[{"id":"gpt-6-luna|codex|low","model":{"label":"GPT-6 Luna"},"accuracy":25.71428571428571,"costPerTaskUsd":0.03371057214285714,"costedTasks":35},{"id":"gpt-6-luna|codex|high","model":{"label":"GPT-6 Luna"},"accuracy":37.142857142857146,"costPerTaskUsd":0.03712029085714286,"costedTasks":35},{"id":"deepseek/deepseek-v4.1-flash-1017|pi|high","model":{"label":"DeepSeek V4.1 Flash 1017"},"accuracy":65.71428571428571,"costPerTaskUsd":0.0782869037142857,"costedTasks":35},{"id":"claude-opus-5-5|claude-code|high","model":{"label":"Claude Opus 5.5"},"accuracy":68.57142857142857,"costPerTaskUsd":2.3308018285714294,"costedTasks":35}],"others":[{"id":"gpt-6-astra|codex|high","model":{"label":"GPT-6 Astra"},"accuracy":65.71428571428571,"costPerTaskUsd":3.183681071428572,"costedTasks":35},{"id":"gpt-6.1-sol|codex|high","model":{"label":"GPT-6.1 Sol"},"accuracy":62.857142857142854,"costPerTaskUsd":0.6359223885714286,"costedTasks":35},{"id":"claude-opus-5-5|claude-code|low","model":{"label":"Claude Opus 5.5"},"accuracy":60,"costPerTaskUsd":0.8323291771428574,"costedTasks":35},{"id":"gpt-6-sol|codex|high","model":{"label":"GPT-6 Sol"},"accuracy":54.285714285714285,"costPerTaskUsd":1.2602519085714288,"costedTasks":35},{"id":"glm-5.3|pi|high","model":{"label":"GLM 5.3"},"accuracy":51.42857142857142,"costPerTaskUsd":3.5096409965714286,"costedTasks":35}],"endorsed":false,"defaultLine":true},{"repository":{"id":873328,"fullName":"getsentry/sentry","description":"Developer-first error tracking and performance monitoring","stars":45591,"ownerAvatarUrl":"https://avatars.githubusercontent.com/u/1396951?v=4"},"publisher":{"kind":"org","login":"mupt-ai","avatarUrl":"https://avatars.githubusercontent.com/u/154857038?v=4"},"releaseId":"ddfe63f0-bc03-4cbe-b879-b92d3f329562","releasedAt":"2026-10-17T18:16:41.795Z","tasks":78,"settings":15,"picks":[{"setting":{"id":"gpt-6-luna|codex|high","model":{"label":"GPT-6 Luna"},"accuracy":51.28205128205128,"costPerTaskUsd":0.02455961493589742,"costedTasks":78},"roles":["cheapest"]},{"setting":{"id":"claude-sonnet-5-5|claude-code|high","model":{"label":"Claude Sonnet 5.5"},"accuracy":82.05128205128204,"costPerTaskUsd":0.5869907666666666,"costedTasks":78},"roles":["mostAccurate"]}],"frontier":[{"id":"gpt-6-luna|codex|high","model":{"label":"GPT-6 Luna"},"accuracy":51.28205128205128,"costPerTaskUsd":0.02455961493589742,"costedTasks":78},{"id":"claude-haiku-5-5|claude-code|low","model":{"label":"Claude Haiku 5.5"},"accuracy":66.66666666666666,"costPerTaskUsd":0.0693667906410256,"costedTasks":78},{"id":"deepseek/deepseek-v4.1-flash-1017|pi|high","model":{"label":"DeepSeek V4.1 Flash 1017"},"accuracy":79.48717948717949,"costPerTaskUsd":0.06940388684615381,"costedTasks":78},{"id":"gpt-6.1-sol|codex|high","model":{"label":"GPT-6.1 Sol"},"accuracy":80.76923076923077,"costPerTaskUsd":0.5029358910256408,"costedTasks":78},{"id":"claude-sonnet-5-5|claude-code|high","model":{"label":"Claude Sonnet 5.5"},"accuracy":82.05128205128204,"costPerTaskUsd":0.5869907666666666,"costedTasks":78}],"others":[{"id":"claude-opus-5-5|claude-code|high","model":{"label":"Claude Opus 5.5"},"accuracy":82.05128205128204,"costPerTaskUsd":1.2209244,"costedTasks":78},{"id":"gpt-6-astra|codex|max","model":{"label":"GPT-6 Astra"},"accuracy":78.2051282051282,"costPerTaskUsd":4.372477717948715,"costedTasks":78},{"id":"claude-opus-5-5|claude-code|low","model":{"label":"Claude Opus 5.5"},"accuracy":76.92307692307693,"costPerTaskUsd":0.530103246153846,"costedTasks":78},{"id":"kimi-k3|pi|high","model":{"label":"Kimi K3"},"accuracy":76.92307692307693,"costPerTaskUsd":1.0855304153846157,"costedTasks":78},{"id":"gpt-6.1-sol|codex|low","model":{"label":"GPT-6.1 Sol"},"accuracy":74.35897435897436,"costPerTaskUsd":0.2737700064102563,"costedTasks":78}],"endorsed":false,"defaultLine":true},{"repository":{"id":1035029907,"fullName":"earendil-works/pi","description":"AI agent toolkit: unified LLM API, agent loop, TUI, coding agent CLI","stars":115162,"ownerAvatarUrl":"https://avatars.githubusercontent.com/u/207902832?v=4"},"publisher":{"kind":"org","login":"mupt-ai","avatarUrl":"https://avatars.githubusercontent.com/u/154857038?v=4"},"releaseId":"06083891-faf0-4b88-8924-32494b7ac170","releasedAt":"2026-10-14T19:55:43.274Z","tasks":47,"settings":15,"picks":[{"setting":{"id":"gpt-6-luna|codex|low","model":{"label":"GPT-6 Luna"},"accuracy":59.57446808510638,"costPerTaskUsd":0.02462187436170212,"costedTasks":47},"roles":["cheapest"]},{"setting":{"id":"custom/dari/routing|pi|default","model":{"label":"dari/routing"},"accuracy":85.1063829787234,"costPerTaskUsd":0.605213906276596,"costedTasks":47},"roles":["mostAccurate"]}],"frontier":[{"id":"gpt-6-luna|codex|low","model":{"label":"GPT-6 Luna"},"accuracy":59.57446808510638,"costPerTaskUsd":0.02462187436170212,"costedTasks":47},{"id":"gpt-6-luna|codex|high","model":{"label":"GPT-6 Luna"},"accuracy":68.08510638297872,"costPerTaskUsd":0.028409147234042552,"costedTasks":47},{"id":"claude-haiku-5-5|claude-code|low","model":{"label":"Claude Haiku 5.5"},"accuracy":78.72340425531915,"costPerTaskUsd":0.1290352240425532,"costedTasks":47},{"id":"gpt-6.1-sol|codex|low","model":{"label":"GPT-6.1 Sol"},"accuracy":82.97872340425532,"costPerTaskUsd":0.36738569574468094,"costedTasks":47},{"id":"custom/dari/routing|pi|default","model":{"label":"dari/routing"},"accuracy":85.1063829787234,"costPerTaskUsd":0.605213906276596,"costedTasks":47}],"others":[{"id":"gpt-6-sol|codex|high","model":{"label":"GPT-6 Sol"},"accuracy":85.1063829787234,"costPerTaskUsd":1.0231192787234045,"costedTasks":47},{"id":"gpt-6.1-sol|codex|high","model":{"label":"GPT-6.1 Sol"},"accuracy":82.97872340425532,"costPerTaskUsd":0.581203644680851,"costedTasks":47},{"id":"typesafe/jev-router|pi|default","model":{"label":"Jev Router"},"accuracy":82.97872340425532,"costPerTaskUsd":0.8114774382978722,"costedTasks":47},{"id":"claude-opus-5-5|claude-code|high","model":{"label":"Claude Opus 5.5"},"accuracy":82.97872340425532,"costPerTaskUsd":1.4360961957446807,"costedTasks":47},{"id":"mistralai/mistral-large-4-0|pi|high","model":{"label":"Mistral Large 4"},"accuracy":80.85106382978722,"costPerTaskUsd":1.5752664110638295,"costedTasks":47}],"endorsed":false,"defaultLine":true},{"repository":{"id":214587193,"fullName":"supabase/supabase","description":"The Postgres development platform. Supabase gives you a dedicated Postgres database to build your web, mobile, and AI applications.","stars":111282,"ownerAvatarUrl":"https://avatars.githubusercontent.com/u/54469796?v=4"},"publisher":{"kind":"org","login":"mupt-ai","avatarUrl":"https://avatars.githubusercontent.com/u/154857038?v=4"},"releaseId":"b5442e9f-8816-49ab-bc48-378bdd85e17f","releasedAt":"2026-10-10T00:13:52.617Z","tasks":36,"settings":12,"picks":[{"setting":{"id":"gpt-6-luna|codex|low","model":{"label":"GPT-6 Luna"},"accuracy":44.44444444444444,"costPerTaskUsd":0.017213391944444448,"costedTasks":36},"roles":["cheapest"]},{"setting":{"id":"gpt-6.1-sol|codex|high","model":{"label":"GPT-6.1 Sol"},"accuracy":77.77777777777779,"costPerTaskUsd":0.4560812666666667,"costedTasks":36},"roles":["mostAccurate"]}],"frontier":[{"id":"gpt-6-luna|codex|low","model":{"label":"GPT-6 Luna"},"accuracy":44.44444444444444,"costPerTaskUsd":0.017213391944444448,"costedTasks":36},{"id":"gpt-6-luna|codex|high","model":{"label":"GPT-6 Luna"},"accuracy":66.66666666666666,"costPerTaskUsd":0.019399179166666662,"costedTasks":36},{"id":"deepseek/deepseek-v4.1-flash|pi|high","model":{"label":"DeepSeek V4.1 Flash"},"accuracy":69.44444444444444,"costPerTaskUsd":0.20034211866666665,"costedTasks":36},{"id":"claude-sonnet-5-5|claude-code|low","model":{"label":"Claude Sonnet 5.5"},"accuracy":72.22222222222221,"costPerTaskUsd":0.20288785555555555,"costedTasks":36},{"id":"gpt-6.1-sol|codex|low","model":{"label":"GPT-6.1 Sol"},"accuracy":75,"costPerTaskUsd":0.23820705555555552,"costedTasks":36},{"id":"gpt-6.1-sol|codex|high","model":{"label":"GPT-6.1 Sol"},"accuracy":77.77777777777779,"costPerTaskUsd":0.4560812666666667,"costedTasks":36}],"others":[{"id":"claude-sonnet-5-5|claude-code|high","model":{"label":"Claude Sonnet 5.5"},"accuracy":77.77777777777779,"costPerTaskUsd":0.4994243500000002,"costedTasks":36},{"id":"kimi-k3|pi|high","model":{"label":"Kimi K3"},"accuracy":77.77777777777779,"costPerTaskUsd":0.7780850666666664,"costedTasks":36},{"id":"claude-opus-5-5|claude-code|high","model":{"label":"Claude Opus 5.5"},"accuracy":77.77777777777779,"costPerTaskUsd":1.2439553111111108,"costedTasks":36},{"id":"claude-opus-5-5|claude-code|low","model":{"label":"Claude Opus 5.5"},"accuracy":75,"costPerTaskUsd":0.36486173333333327,"costedTasks":36},{"id":"z-ai/glm-5.3-flash|pi|high","model":{"label":"GLM 5.3 Flash"},"accuracy":66.66666666666666,"costPerTaskUsd":0.10426330083333332,"costedTasks":36}],"endorsed":false,"defaultLine":true},{"repository":{"id":1271113810,"fullName":"vercel/eve","description":"The Open Framework for Building Agents","stars":5448,"ownerAvatarUrl":"https://avatars.githubusercontent.com/u/14985020?v=4"},"publisher":{"kind":"org","login":"mupt-ai","avatarUrl":"https://avatars.githubusercontent.com/u/154857038?v=4"},"releaseId":"f2164d5e-59e7-4a5c-b42c-75f57ce74a9a","releasedAt":"2026-10-02T23:55:44.395Z","tasks":38,"settings":10,"picks":[{"setting":{"id":"gpt-6-luna|codex|high","model":{"label":"GPT-6 Luna"},"accuracy":50,"costPerTaskUsd":0.0404098152631579,"costedTasks":38},"roles":["cheapest"]},{"setting":{"id":"gpt-6.1-sol|codex|low","model":{"label":"GPT-6.1 Sol"},"accuracy":73.68421052631578,"costPerTaskUsd":0.42938203421052634,"costedTasks":38},"roles":["mostAccurate"]}],"frontier":[{"id":"gpt-6-luna|codex|high","model":{"label":"GPT-6 Luna"},"accuracy":50,"costPerTaskUsd":0.0404098152631579,"costedTasks":38},{"id":"gpt-6.1-sol|codex|low","model":{"label":"GPT-6.1 Sol"},"accuracy":73.68421052631578,"costPerTaskUsd":0.42938203421052634,"costedTasks":38}],"others":[{"id":"claude-opus-5-5|claude-code|high","model":{"label":"Claude Opus 5.5"},"accuracy":73.68421052631578,"costPerTaskUsd":2.1562288210526317,"costedTasks":38},{"id":"gpt-6.1-sol|codex|high","model":{"label":"GPT-6.1 Sol"},"accuracy":71.05263157894737,"costPerTaskUsd":0.8453347789473684,"costedTasks":38},{"id":"claude-opus-5-5|claude-code|low","model":{"label":"Claude Opus 5.5"},"accuracy":71.05263157894737,"costPerTaskUsd":0.8460090894736844,"costedTasks":38},{"id":"kimi-k3|pi|high","model":{"label":"Kimi K3"},"accuracy":65.78947368421053,"costPerTaskUsd":2.646881526315789,"costedTasks":38},{"id":"deepseek/deepseek-v4.1-flash|pi|high","model":{"label":"DeepSeek V4.1 Flash"},"accuracy":60.526315789473685,"costPerTaskUsd":0.4305551952631579,"costedTasks":38}],"endorsed":false,"defaultLine":true},{"repository":{"id":67753070,"fullName":"vercel/vercel","description":"Develop. Preview. Ship.","stars":16336,"ownerAvatarUrl":"https://avatars.githubusercontent.com/u/14985020?v=4"},"publisher":{"kind":"org","login":"mupt-ai","avatarUrl":"https://avatars.githubusercontent.com/u/154857038?v=4"},"releaseId":"99f50904-1e3d-45b6-b3a5-0ffd8677a75a","releasedAt":"2026-10-02T22:39:07.640Z","tasks":47,"settings":8,"picks":[{"setting":{"id":"gpt-6-luna|codex|high","model":{"label":"GPT-6 Luna"},"accuracy":68.08510638297872,"costPerTaskUsd":0.01846890276595745,"costedTasks":47},"roles":["cheapest"]},{"setting":{"id":"claude-opus-5-5|claude-code|high","model":{"label":"Claude Opus 5.5"},"accuracy":80.85106382978722,"costPerTaskUsd":1.2726437872340426,"costedTasks":47},"roles":["mostAccurate"]}],"frontier":[{"id":"gpt-6-luna|codex|high","model":{"label":"GPT-6 Luna"},"accuracy":68.08510638297872,"costPerTaskUsd":0.01846890276595745,"costedTasks":47},{"id":"z-ai/glm-5.3-flash|pi|high","model":{"label":"GLM 5.3 Flash"},"accuracy":72.3404255319149,"costPerTaskUsd":0.08569868957446805,"costedTasks":47},{"id":"gpt-6.1-sol|codex|low","model":{"label":"GPT-6.1 Sol"},"accuracy":76.59574468085107,"costPerTaskUsd":0.2219091042553192,"costedTasks":47},{"id":"kimi-k3|pi|high","model":{"label":"Kimi K3"},"accuracy":78.72340425531915,"costPerTaskUsd":0.9614162680851065,"costedTasks":47},{"id":"claude-opus-5-5|claude-code|high","model":{"label":"Claude Opus 5.5"},"accuracy":80.85106382978722,"costPerTaskUsd":1.2726437872340426,"costedTasks":47}],"others":[{"id":"gpt-6.1-sol|codex|high","model":{"label":"GPT-6.1 Sol"},"accuracy":72.3404255319149,"costPerTaskUsd":0.4103219787234043,"costedTasks":47},{"id":"glm-5.3|pi|high","model":{"label":"GLM 5.3"},"accuracy":70.2127659574468,"costPerTaskUsd":1.2834871421276595,"costedTasks":47},{"id":"deepseek/deepseek-v4.1-flash|pi|high","model":{"label":"DeepSeek V4.1 Flash"},"accuracy":63.829787234042556,"costPerTaskUsd":0.22483220519148928,"costedTasks":47}],"endorsed":false,"defaultLine":true},{"repository":{"id":1153130349,"fullName":"pingdotgg/t3code","stars":24306,"ownerAvatarUrl":"https://avatars.githubusercontent.com/u/89191727?v=4"},"publisher":{"kind":"org","login":"mupt-ai","avatarUrl":"https://avatars.githubusercontent.com/u/154857038?v=4"},"releaseId":"46a4c74e-b035-4613-b0c2-2637e9a766f1","releasedAt":"2026-10-02T22:36:08.549Z","tasks":39,"settings":9,"picks":[{"setting":{"id":"gpt-6-luna|codex|high","model":{"label":"GPT-6 Luna"},"accuracy":46.15384615384615,"costPerTaskUsd":0.037430796282051274,"costedTasks":39},"roles":["cheapest"]},{"setting":{"id":"claude-opus-5-5|claude-code|high","model":{"label":"Claude Opus 5.5"},"accuracy":76.92307692307693,"costPerTaskUsd":2.051387333333333,"costedTasks":39},"roles":["mostAccurate"]}],"frontier":[{"id":"gpt-6-luna|codex|high","model":{"label":"GPT-6 Luna"},"accuracy":46.15384615384615,"costPerTaskUsd":0.037430796282051274,"costedTasks":39},{"id":"gpt-6.1-sol|codex|low","model":{"label":"GPT-6.1 Sol"},"accuracy":66.66666666666666,"costPerTaskUsd":0.3065382,"costedTasks":39},{"id":"claude-opus-5-5|claude-code|high","model":{"label":"Claude Opus 5.5"},"accuracy":76.92307692307693,"costPerTaskUsd":2.051387333333333,"costedTasks":39}],"others":[{"id":"z-ai/glm-5.3-flash|pi|high","model":{"label":"GLM 5.3 Flash"},"accuracy":64.1025641025641,"costPerTaskUsd":0.3535161264102564,"costedTasks":39},{"id":"gpt-6.1-sol|codex|high","model":{"label":"GPT-6.1 Sol"},"accuracy":64.1025641025641,"costPerTaskUsd":0.5991060230769232,"costedTasks":39},{"id":"claude-sonnet-5-5|claude-code|high","model":{"label":"Claude Sonnet 5.5"},"accuracy":58.97435897435898,"costPerTaskUsd":0.874405366666667,"costedTasks":39},{"id":"kimi-k3|pi|high","model":{"label":"Kimi K3"},"accuracy":56.41025641025641,"costPerTaskUsd":2.379947061538461,"costedTasks":39},{"id":"glm-5.3|codex|high","model":{"label":"GLM 5.3"},"accuracy":51.28205128205128,"costPerTaskUsd":3.007110388717948,"costedTasks":39}],"endorsed":false,"defaultLine":true}]}