Qwen3.6 35B A3B API

Qwen3.6 35B A3B is a 256-expert mixture-of-experts reasoning model with 128K context, function tools, and strict structured JSON output.

Alibaba CloudGénération de texte128K contextePublié 29 juil. 2026Inférence nativeNouveau

À propos de Qwen3.6 35B A3B

Qwen3.6 35B A3B is a 256-expert mixture-of-experts reasoning model with 128K context, function tools, and strict structured JSON output.

Text-only. This build does not accept image or video input, unlike the base Qwen3.6 35B A3B. Weights Served from the 2-bit eschamoe W2 build published by Escha Labs (eschalabs.com) as EschaLabs/Qwen3.6-35B-A3B-Escha-W2 on Hugging Face, under Apache-2.0. The experts are quantized to 2 bits, mixed per projection (gate_up_proj at 2-bit and down_proj at 3-bit), the dense layers are int8, and the KV cache is FP16. Escha Labs publishes the quality comparison against an FP8 baseline of the same model: parity or better on math, graduate science, tool use and long context, about 2 percent lower on broad knowledge, and about 7 percent lower on long-horizon code generation, which is the one clear gap. See the model card for the full benchmark table and protocol. Behavior Supports streaming, function tools, structured JSON output including strict schemas, and thinking mode on by default. Set enable_thinking=false for direct answers. With thinking on, the reasoning arrives in reasoning_content and the answer in content, so read both. A low max_tokens with thinking on can be spent entirely on reasoning, so allow room for the answer. Caching Automatic prefix cache reads are billed at the cached-input rate when reported. Explicit cache controls are not supported. Cancelling a streaming request mid-generation bills only the tokens produced up to that point.

Aussi connu sous le nom Alibaba Cloud Qwen3.6 35B A3B, Qwen3.6-35B-A3B, qwen3-6-35b-a3b

reasoningfunction callingjson modecache

Caractéristiques de Qwen3.6 35B A3B

ID du modèle
qwen3-6-35b-a3b
Fournisseur
Alibaba Cloud
Catégorie
Génération de texte
Publié
29 juil. 2026
Fenêtre de contexte
128K tokens
Sortie max
16 384 tokens
Entrée
Texte
Sortie
Texte
Sortie structurée
JSON Schema
Endpoints
POST/v1/chat/completionsPOST/v1/responsesPOST/v1/messagesPOST/v1/completionsPOST/v1beta/models/qwen3-6-35b-a3b:generateContent
Identifiants de modèle alternatifs
qwen3.6-35b-a3bqwen/qwen3.6-35b-a3bEschaLabs/Qwen3.6-35B-A3B-Escha-W2

Tarifs de l'API Qwen3.6 35B A3BÉconomisez jusqu'à 72%

Tarifs à l'usage en direct du catalogue EmpirioLabs. Vous ne payez que ce que vous utilisez, sans minimum mensuel.

Type
Spéc.
Tarif
Entrée
per 1M prompt tokens
$0.248$0.07
Sortie
per 1M generated tokens
$1.485$0.42
Implicit cache read
per 1M cached input tokens
$0.035
Web Search (Linkup)
per call when invoked
$0.013
Comparer sur la page complète des tarifs

Comment appeler l'API Qwen3.6 35B A3B

Qwen3.6 35B A3B sert l'API Chat Completions compatible OpenAI. Pointez n'importe quel SDK OpenAI vers https://api.empiriolabs.ai/v1 avec votre clé API EmpirioLabs et utilisez l'id de modèle qwen3-6-35b-a3b. Obtenez une clé API depuis le tableau de bord EmpirioLabs.

cURL
curl https://api.empiriolabs.ai/v1/chat/completions \
  -H "Authorization: Bearer $EMPIRIOLABS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3-6-35b-a3b",
    "messages": [
      {"role": "user", "content": "Write a haiku about the ocean."}
    ]
  }'
Python (OpenAI SDK)
from openai import OpenAI

client = OpenAI(
    base_url="https://api.empiriolabs.ai/v1",
    api_key="YOUR_EMPIRIOLABS_API_KEY",
)

response = client.chat.completions.create(
    model="qwen3-6-35b-a3b",
    messages=[{"role": "user", "content": "Write a haiku about the ocean."}],
)
print(response.choices[0].message.content)
Référence complète de l'API Qwen3.6 35B A3B

Paramètres de l'API Qwen3.6 35B A3B

Paramètres de requête pris en charge par l'API Qwen3.6 35B A3B sur EmpirioLabs. Les valeurs par défaut s'appliquent quand un champ est omis.

ParamètreTypePar défautPlage / valeursDescription
temperaturenumber0.70 à 2Sampling temperature. 0 is deterministic and 2 is maximum randomness.
top_pnumber0.950 à 1Nucleus sampling probability mass. Lower values make outputs more focused.
max_tokensnumber40961 à 16384Maximum output tokens. With thinking on, leave room for the answer as well as the reasoning.
stopstring--Up to 4 strings where the model will stop generating further tokens.
enable_thinkingbooleantrue-Enable the model reasoning channel before final output. Reasoning is returned in reasoning_content and bills as output tokens.
reasoning_effortenummediumnone, low, medium, high, maxReasoning effort. none disables thinking. Any other value enables it.
top_knumber201 à 200Limit sampling to the top K candidate tokens.
min_pnumber00 à 1Minimum probability threshold for token sampling.
presence_penaltynumber0-2 à 2Penalize tokens that have already appeared, increasing topic diversity.
frequency_penaltynumber0-2 à 2Penalize tokens in proportion to how often they have appeared.
seednumber-0 à 2147483647Best-effort determinism. Output is not bit-reproducible across requests because batch composition changes accumulation order.
response_formatenum-text, json_object, json_schemaStructured output. json_object returns valid JSON. json_schema enforces your schema exactly.
web_search_linkupbooleanfalse-Optional web search powered by Linkup. When enabled, recent web sources are retrieved using your latest user message as the query and provided to the model as additional context. Adds $0.013 per call when invoked on top of the model's normal token cost. Disabled by default.

Bon à savoir

Text-only. This build does not accept image or video input, unlike the base Qwen3.6 35B A3B.

Weights

Served from the 2-bit eschamoe W2 build published by Escha Labs (eschalabs.com) as EschaLabs/Qwen3.6-35B-A3B-Escha-W2 on Hugging Face, under Apache-2.0. The experts are quantized to 2 bits, mixed per projection (gate_up_proj at 2-bit and down_proj at 3-bit), the dense layers are int8, and the KV cache is FP16. Escha Labs publishes the quality comparison against an FP8 baseline of the same model: parity or better on math, graduate science, tool use and long context, about 2 percent lower on broad knowledge, and about 7 percent lower on long-horizon code generation, which is the one clear gap. See the model card for the full benchmark table and protocol.

Behavior

Supports streaming, function tools, structured JSON output including strict schemas, and thinking mode on by default. Set enable_thinking=false for direct answers. With thinking on, the reasoning arrives in reasoning_content and the answer in content, so read both. A low max_tokens with thinking on can be spent entirely on reasoning, so allow room for the answer.

Caching

Automatic prefix cache reads are billed at the cached-input rate when reported. Explicit cache controls are not supported. Cancelling a streaming request mid-generation bills only the tokens produced up to that point.

API Qwen3.6 35B A3B : questions fréquentes

Combien coûte l'API Qwen3.6 35B A3B ?

Sur EmpirioLabs, Qwen3.6 35B A3B est facturé à l'usage. La grille tarifaire en direct de cette page correspond toujours à ce que l'API facture.

Quelle est la fenêtre de contexte de Qwen3.6 35B A3B ?

Qwen3.6 35B A3B prend en charge une fenêtre de contexte de 128K tokens avec jusqu'à 16 384 tokens de sortie par réponse.

L'API Qwen3.6 35B A3B est-elle compatible OpenAI ?

Oui. Qwen3.6 35B A3B sert l'API Chat Completions compatible OpenAI : les SDKs OpenAI existants fonctionnent en pointant base_url vers https://api.empiriolabs.ai/v1 et en utilisant l'id de modèle qwen3-6-35b-a3b.

Puis-je essayer Qwen3.6 35B A3B dans le navigateur avant d'intégrer ?

Oui. Le playground EmpirioLabs exécute Qwen3.6 35B A3B dans le navigateur avec les mêmes paramètres que l'API expose, pour tester vos prompts avant d'écrire du code.

Comment obtenir une clé API Qwen3.6 35B A3B ?

Créez un compte EmpirioLabs, puis générez une clé sous API Keys dans le tableau de bord. La facturation utilise des crédits à l'usage : vous ne payez que vos requêtes.

Prêt à utiliser de meilleurs points de terminaison ?

Consultez nos tarifs ou contactez-nous si vous souhaitez que votre propre modèle soit déployé sur notre stack.