Recursive: Flint (free)

{"about":"### 用於程式碼工作的模型\n\nFlint 是 Recursive 為程式代理打造的文字模型，用於了解程式碼庫、修正錯誤、實作功能及撰寫測試。它是公開權重的 MoE 模型：總計 398 億參數，每個 token 啟用約 25 億參數。專家分工屬於模型架構，不代表聊天中有多個獨立助理。\n\nRecursive 的研究系統 Eureka Machine 參與了 Flint 的開發，協助尋找架構方案、選擇資料與訓練方法，以及最佳化運算。Flint 本身經過預訓練、中期訓練、指示範例訓練與強化學習。訓練工作包含文字、程式碼、遵循指示及代理情境。\n\n### 分析專案與準備修改\n\n實際工作可從專案的特定部分開始：以文字提供必要檔案，描述函式用途與預期結果。尋找錯誤時，加入觀察到的行為與日誌片段，再要求解釋原因、提供最小修正，以及能區分正確行為與錯誤的測試。這種順序便於後續在自己的環境驗證。\n\n新功能請指定輸入資料、限制及結果範例。重構時列出必須保留的特性，並要求另行說明介面修改。開發商建議將複雜工作拆成小型子任務並給出明確指示。因此，包含中間結果的連續對話，比籠統要求重做整個專案更符合 Flint 的預期用法。\n\n### 上下文與連線格式\n\n在 Ailoot，Flint 的輸入與輸出都是文字。上下文視窗為 262,144 個 token，原始資料、對話歷史與答案共用此空間。輸出上限標示為 235,929 個 token。這些數值描述連線界線，實際長度宜依任務指定，例如錯誤原因說明、修補內容、測試組或工作計畫。\n\n可使用 Messages、Responses 與 Chat Completions 格式，便於沿用客戶端應用程式的常見結構。本機檔案讀取及修改套用由應用程式自行安排；在文字請求中，模型收到的是傳入程式碼，回傳的是建議、解釋或新程式碼。\n\nFlint 適合反覆進行技術對話，每個後續步驟都根據專案內容及前一次修改的驗證結果。它的專長有助於表達任務，實際成果則是具體文字：說明、實作或可驗證的測試。","alias_of":null,"aliases":[],"author":"Recursive","canonical_model":"recursive/flint-v1.0:free","ctx":262144,"formats":["chat","messages","responses"],"id":"recursive/flint-v1.0:free","input":["text"],"kind":"text","max_out":235929,"methods":[{"cache_read":null,"cache_write":null,"cache_write_1h":null,"cache_write_5m":null,"conditional_meter":null,"ctx":262144,"family":"anthropic","fast_multiplier":1,"flex_multiplier":1,"fmt":"messages","http":"POST","img_in":null,"img_out":null,"in":0,"input":["text"],"label":"Anthropic Messages","max_out":235929,"max_tool_calls":null,"out":0,"output":["text"],"path":"/v1/messages","request_model_id":"recursive/flint-v1.0:free","route_label":"Ultra","status_reads":[],"tiers":[],"tool_prices":null,"unit":"token","units":[]},{"cache_read":null,"cache_write":null,"cache_write_1h":null,"cache_write_5m":null,"conditional_meter":null,"ctx":262144,"family":"openai","fast_multiplier":1,"flex_multiplier":1,"fmt":"responses","http":"POST","img_in":null,"img_out":null,"in":0,"input":["text"],"label":"OpenAI Responses","max_out":235929,"max_tool_calls":null,"out":0,"output":["text"],"path":"/v1/responses","request_model_id":"recursive/flint-v1.0:free","route_label":"Ultra","status_reads":[],"tiers":[],"tool_prices":null,"unit":"token","units":[]},{"cache_read":null,"cache_write":null,"cache_write_1h":null,"cache_write_5m":null,"conditional_meter":null,"ctx":262144,"family":"openai","fast_multiplier":1,"flex_multiplier":1,"fmt":"chat","http":"POST","img_in":null,"img_out":null,"in":0,"input":["text"],"label":"OpenAI Chat","max_out":235929,"max_tool_calls":null,"out":0,"output":["text"],"path":"/v1/chat/completions","request_model_id":"recursive/flint-v1.0:free","route_label":"Ultra","status_reads":[],"tiers":[],"tool_prices":null,"unit":"token","units":[]}],"model_aliases":[{"canonical_model_id":"recursive/flint-v1.0:free","request_model_id":"recursive/flint-v1.0:free","route_label":"Ultra"}],"name":"Recursive: Flint (free)","price":{"in":0,"in_label":null,"out":0,"out_label":null,"sort_in":0,"sort_out":0,"varies":false},"public_id":"recursive/flint-v1.0:free","short":"Recursive 的文字 MoE 模型，用於了解程式碼庫、修正錯誤，以及準備功能與測試。"}