Canji

Request mai web_search: true yana ambaton tushensa: sources a /v1/chat/completions da /v1/messages yana jera sakamakon da aka ba model, kuma a /v1/chat/completions annotations yana nuna inda amsar ta ambaci kowanne. Binciken Yanar Gizon da aka Gina

web_search: true a /v1/chat/completions da /v1/messages yana gudanar da binciken yanar gizo kafin model ya amsa kuma yana ba shi sakamakon a matsayin tushe masu lamba. Request da ya sami sakamako yana amfani da bincike ɗaya na allowance ɗinku na yau da kullun. Idan babu sauran, ana amsa request ba tare da bincike ba. Binciken Yanar Gizon da aka Gina

Stream da aka yanke a gefen server ana cajin sa don usage da aka bayyana har wannan lokacin. Sauran abin da aka ware yana komawa balance ɗinku.

shannon-coder-1 yana amsawa a /v1/responses kamar yadda yake yi a sauran endpoints biyu. Kayan CLI na rubuta code

Maɓallin API da aka canza yana daina aiki nan take.

POST /v1/tokenize da POST /v1/messages/count_tokens: ƙirgan tokens kyauta ga hosted open-weight models. Ƙidayar tokens

Adana prompt a hosted open-weight models. Ana cajin cached input a 25% na farashin input kuma ana bayyana shi a usage a dukkan endpoints uku: prompt_tokens_details.cached_tokens, input_tokens_details.cached_tokens da cache_read_input_tokens. GET /v1/models yana bayyana pricing.cached_input_per_million_usd da capabilities.prompt_caching ga waɗannan models. Adana prompt

Hosted open-weight models: tool_choice na tilas (required ko function mai suna) yana samar da kiran tool. Kiran Aiki

Hosted open-weight models suna karɓar saƙon system wanda ba shi ne saƙon farko na tattaunawa ba.

Ana iya isa API a matsayin Tor onion service da kuma a adireshin I2P. Amsoshin api.shannon-ai.com suna ɗauke da adireshin biyu a headers Onion-Location da X-I2P-Location. Endpoints masu ɓoyewa, Tor & I2P

Sabbin models shannon-3.1 da shannon-3.1-pro, da GET /v1/models ke lissafawa. Models & farashi

Request da aka yi streaming wanda client ɗinsa ya katse yana ci gaba har ƙarshe kuma ana ƙirga shi. A da, amsar tana tsayawa a lokacin katsewa.

Requests zuwa /v1/responses suna bayyana a usage log.

Hosted open-weight models suna aiwatar da sampling fields: temperature, top_p, seed, stop da max_tokens a /v1/chat/completions; temperature, top_p da stop_sequences a /v1/messages; temperature, top_p da max_output_tokens a /v1/responses. Chat Completions

Usage log yana nuna kuɗi a USD ga kowane kira da kuma jimilla.

Ƙimar model da ba id da aka wallafa ba ana amsa ta da 400, nau'in invalid_request_error da saƙon unknown model: <id> a dukkan endpoints uku. A da, shannon-1.6-lite ne yake amsa irin wannan request. Models & farashi

/v1 yana karɓar maɓallin API kaɗai. Ba a karanta cookie na shiga na gidan yanar gizo a can. Tabbatarwa

Flood protection: requests 120 a minti ɗaya ga kowane asusu a endpoints uku na samarwa. Iyakoki da balance

Idan model ya kasa amsawa, kuskuren yana ɗauke da tsayayyen saƙo The model backend failed to answer. Please retry. Duba balance da ba za a iya yi ba ana amsa shi da 503 da nau'in api_error. Kuskuren Gudanarwa

Request da ba a yi streaming ba kuma ya gaza yana mayar da abin da aka ware zuwa balance ɗinku.

GET /v1/models yana bayyana capabilities.reasoning_effort ga kowane model. Reasoning effort

Hosted open-weight models: request da ba a yi streaming ba wanda bai samu amsa ba yana ƙarewa da 504 da nau'in api_error. Sakamako mara komai shi ne 502. Kuskuren Gudanarwa

Hosted open-weight models suna karɓar input da model mai suna yake karɓa. Hoto da aka aika zuwa id na rubutu kaɗai ana amsa shi da 400 da <id> does not accept image input. response_format a kan id ba tare da structured output ba ana amsa shi da 400. GET /v1/models yana bayyana capabilities.vision, response_format da json_schema ga kowane id. Hotuna da fayiloli

Amsoshin da ba a yi streaming ba suna ɗauke da rubutun reasoning: reasoning_content a saƙon /v1/chat/completions, da block na thinking kafin rubutu a /v1/messages. Reasoning effort

Hosted open-weight models suna da sabbin farashi, daga $0.50 kowace 1M input tokens da $2.00 kowace 1M output tokens. Models & farashi

Hosted open-weight models suna da id ɗaya ga kowane model, ids goma sha biyu gaba ɗaya. Models & farashi

Structured output a shannon-3, shannon-3-pro, shannon-3.1 da shannon-3.1-pro. GET /v1/models yana bayyana capabilities.json_schema: true a gare su. Abubuwan da aka Tsara

/v1/responses yana karanta text.format (json_object ko json_schema) don structured output. Responses API

/v1/responses yana mayar da reasoning na models na Shannon: item na output reasoning kafin item na message, da stream events response.reasoning_summary_text.delta da response.reasoning_summary_text.done. Responses API

Field ɗin model na amsa shi ne id da aka wallafa a kowane endpoint. Object na Responses yana ɗauke da model da created_at.

GET /v1/models yana bayyana context window na kowane model. Models & farashi

Hosted open-weight models suna amsawa a /v1/messages da /v1/responses ban da /v1/chat/completions, da system prompt ɗinku da kiran tool. /v1/messages yana karɓar thinking, kuma reasoning_effort field ne na request. Models & farashi

Jikunan kurakurai a /v1 suna bin siffofin SDK: object ɗin error mai type da message a endpoints na tsarin OpenAI, da object iri ɗaya a cikin {"type": "error"} a /v1/messages. Kuskuren Gudanarwa

Keep-alive a buɗaɗɗun streams kowane daƙiƙa 15: layin sharhi : ping a /v1/chat/completions, da event: ping a /v1/messages. Yawo

Ids na kiran tool suna farawa da call_ a /v1/chat/completions da /v1/responses, da toolu_ a /v1/messages. Kiran Aiki

/v1/responses: stream na asali shi ne false. Responses API

Ana karɓar max_completion_tokens a matsayin wani suna na max_tokens a /v1/chat/completions. Chat Completions

Event response.completed na Responses stream yana ɗauke da array ɗin output. Responses API

Shigarwar GET /v1/models tana ɗauke da display_name, context_window, max_output_tokens, pricing, capabilities da endpoints. Models & farashi

Kowace amsa tana ɗauke da x-request-id, kuma ƙimar da kuka aika ana mayar da ita. Jikunan requests na iya kaiwa 32 MiB.

Ana iya soke maɓalli ba tare da an ba da sabo ba. An raba usage log zuwa shafuka.

Hotuna da takardu a requests: sassan image_url a /v1/chat/completions, blocks na image, document da file a /v1/messages, sassan input_image a /v1/responses. Hotuna da fayiloli

Sabon model shannon-3-pro. Models & farashi

Sabon model shannon-3. Models & farashi

Streams a /v1/chat/completions suna ɗauke da reasoning a matsayin delta.reasoning_content. Reasoning effort

Ana riƙe max_tokens tsakanin 1 zuwa 65,536. Ƙimar da ta wuce wannan kewayon ana kawo ta zuwa mafi kusa, ba a ƙi ta ba. Chat Completions

Ana ƙirga shannon-coder-1 a kira ga kowane plan a /v1/chat/completions da /v1/messages. Idan an gama kiran taga, amsar 429 ce da saƙon Shannon Coder call quota reached. Kayan CLI na rubuta code

Tattaunawar da ta fi context window girma ana amsa ta da 400, nau'in invalid_request_error da code context_length_exceeded. Kuskuren da ya faru a cikin stream yana zuwa a matsayin error event, ba a matsayin rubutu a cikin amsar ba. Kayan CLI na rubuta code

Endpoints na /v1: POST /v1/chat/completions, POST /v1/messages, POST /v1/responses da GET /v1/models. Su ne endpoints da OpenAI da Anthropic SDKs, Claude Code da Codex CLI ke kira. Dubawa

Kiran tool da tools a /v1/chat/completions da /v1/messages, structured output da response_format, da system a matsayin string ko text blocks a /v1/messages. Kiran Aiki

Model ids shannon-1.6-pro, shannon-2-lite, shannon-2-pro da shannon-coder-1, kusa da shannon-1.6-lite. Models & farashi

Fitowar API ta farko: taɗi a tsarin OpenAI da tsarin Anthropic, da endpoint na Responses, kowanne ana kiransa da maɓallin API da ya fara da sk-. Tabbatarwa