Dumphani kupita ku zomwe zili
Mphamvu

Mphamvu

Zomwe model iliyonse imatha kudzera mu API, model ndi model: ma tool, input ya zithunzi, reasoning, JSON output, prompt caching, web search ndi streaming.

Model iliyonse imafikiridwa kudzera mu ma endpoint atatu omwewo ndi kiyi yomweyo. Ma table awiri amaonetsa komwe ma model amasiyana. Dash imatanthauza kuti model imayankha popanda luso limenelo.

Ma model a Shannon

Model Ma toolZithunziReasoningEffortJSON objectJSON schemaCachingKusakaStreaming
shannon-1.6-lite IndeInde——IndeInde—IndeInde
shannon-1.6-pro IndeIndeInde—IndeInde—IndeInde
shannon-2-lite Inde—Inde—IndeInde—IndeInde
shannon-2-pro Inde—Inde—IndeInde—IndeInde
shannon-3 IndeIndeInde—IndeInde—IndeInde
shannon-3-pro IndeIndeInde—IndeInde—IndeInde
shannon-3.1 IndeIndeInde—IndeInde—IndeInde
shannon-3.1-pro IndeIndeInde—IndeInde—IndeInde
shannon-3.5 Inde———IndeInde——Inde
shannon-coder-1 Inde—Inde—IndeInde——Inde

Hosted open-weight model

Id ya hosted model ndi dzina lake ndi variant yake zolumikizidwa ndi hyphen, mwachitsanzo DeepSeek-V4-Pro-0813-3BIT-REAP.

Model Ma toolZithunziReasoningEffortJSON objectJSON schemaCachingKusakaStreaming
DeepSeek-V4-Pro-0813 3BIT-REAP Inde—IndeIndeIndeIndeInde—Inde
GLM-5.2 3BIT-REAP Inde—IndeIndeIndeIndeInde—Inde
Kimi-K3 3BIT-REAP IndeIndeIndeIndeIndeIndeInde—Inde
Nemotron3Ultra 3BIT-REAP Inde—IndeIndeIndeIndeInde—Inde
MiniMax-M3 3BIT-REAP IndeIndeIndeIndeIndeIndeInde—Inde
DeepSeek-V4-Flash-0731 W4A16-AUTOROUND-REAP Inde—IndeIndeIndeIndeInde—Inde
Kimi-K2.6 W4A16-AUTOROUND-REAP IndeIndeIndeIndeIndeIndeInde—Inde
Laguna-S-2.1 W4A16-AUTOROUND-REAP Inde—IndeInde——Inde—Inde
inkling W4A16-AUTOROUND-REAP IndeIndeIndeIndeInde—Inde—Inde
MiMo-V2.5-Pro W8A16 Inde—IndeIndeIndeIndeInde—Inde
MiMo-V2.5 W8A16 IndeIndeIndeIndeIndeIndeInde—Inde
Hy3 W8A16 Inde—IndeIndeIndeIndeInde—Inde

Tanthauzo la column iliyonse

  • Ma tool. Mumafotokoza ma function mu tools, model imayankha ndi ma call omwe ikufuna kuti achitidwe, ndipo code yanu imawayendetsa. Model iliyonse imalandira ma tool. tool_choice yokakamiza imagwiritsidwa ntchito ndi hosted open-weight model. Kuyitana mafunkseni
  • Input ya zithunzi. Model imawerenga zithunzi zotumizidwa mkati mwa message ya user, ngati data URL kapena ngati adilesi. shannon-2-lite ndi shannon-2-pro zimawerenga zithunzi m'ma pempho omwe amatumizanso tools kapena response_format. Banja la Shannon 3 limawerenganso zikalata. Zithunzi ndi ma file
  • Reasoning. Yankho likhoza kukhala ndi text ya reasoning ya model pambali pa yankho, mu field yake. Reasoning effort
  • Reasoning effort. Field ya pempho imakhazikitsa kuzama kwa reasoning ya model: off, low, medium kapena high. Ma model a Shannon amadzikhazikitsira okha kuzama kwake. Reasoning effort
  • JSON object. Ndi response_format: {"type": "json_object"} yankho limakhala JSON yopanda mawonekedwe operekedwa. Pa /v1/chat/completions ndi /v1/responses. Zotuluka zokonzedwa
  • JSON schema. Ndi response_format ya mtundu wa json_schema yankho limatsatira schema yomwe mumatumiza. Pa /v1/chat/completions ndi /v1/responses. Zotuluka zokonzedwa
  • Prompt caching. Prompt yoyambira ngati yaposachedwa pa model yomweyo imawerengedwa kuchokera ku cache, ndipo gawo losungidwa limaperekedwa mtengo wa 25% ya mtengo wa input. Siifuna field iliyonse mu pempho. Kusunga prompt
  • Web search. Ndi web_search: true API imasaka pa web ndikupatsa model zotsatira isanayankhe. Pa /v1/chat/completions ndi /v1/messages. Ndi field ya pempho ndipo ilibe flag mu GET /v1/models. Kusaka pa web
  • Streaming. Ndi stream: true yankho limafika ngati server-sent events pamene model ikulemba. Model iliyonse imachita stream pa ma endpoint onse atatu. Kutsitsa nthawi yeniyeni

Mfundo zomwezo kuchokera ku GET /v1/models

GET https://api.shannon-ai.com/v1/models

GET /v1/models imalemba model iliyonse ndi object ya capabilities, kotero pulogalamu imatha kuwerenga mfundozi m'malo mwa tsamba ili. Endpoint iyi siifuna kiyi.

from openai import OpenAI

client = OpenAI(api_key="YOUR_API_KEY", base_url="https://api.shannon-ai.com/v1")

for model in client.models.list():
    print(model.id, getattr(model, "capabilities", None))

id ndi capabilities za ma entry awiri a data:

200 JSON
[
  {
    "id": "shannon-3",
    "capabilities": {
      "tools": true,
      "vision": true,
      "reasoning": true,
      "reasoning_effort": false,
      "streaming": true,
      "json_schema": true
    }
  },
  {
    "id": "Kimi-K3-3BIT-REAP",
    "capabilities": {
      "tools": true,
      "vision": true,
      "reasoning": true,
      "reasoning_effort": true,
      "streaming": true,
      "response_format": true,
      "json_schema": true,
      "prompt_caching": true
    }
  }
]
Column Flag mu capabilities
Ma tool tools
Input ya zithunzi vision
Reasoning reasoning
Reasoning effort reasoning_effort
JSON object response_format
JSON schema json_schema
Prompt caching prompt_caching
Web search Si flag: field ya pempho web_search.
Streaming streaming
  • response_format ndi prompt_caching ndi ma flag a ma entry a hosted open-weight. Entry ya Shannon imanena JSON output ndi json_schema yokha, yomwe imaimira ma JSON mode onse awiri.
  • vision ndi true pa shannon-2-lite ndi shannon-2-pro: zimawerenga zithunzi m'ma pempho omwe amatumizanso tools kapena response_format.

Kunja kwa mzere wa model

Hosted open-weight model imakana chithunzi kapena JSON mode yomwe ili kunja kwa mzere wake. Kukanako kumachitika asanachotse chilichonse pa balance yanu:

Status Mtundu Message Liti
400 invalid_request_error <id> does not accept image input Chithunzi chinatumizidwa kwa hosted open-weight model yopanda input ya zithunzi.
400 invalid_request_error <id> does not accept response_format JSON output idapemphedwa kwa hosted open-weight model yomwe ilibe JSON mode iliyonse.
400 invalid_request_error <id> does not accept response_format json_schema; use json_object Schema inatumizidwa kwa hosted open-weight model yomwe ili ndi JSON object mode yokha.

Model ya Shannon imayankha pempho loterolo ndi status 200: luso lomwe lili kunja kwa mzere wake silikhala ndi gawo pa yankho. Zolakwika