Alibaba: qwen-3-235b-a22b-instruct-2507-tput:free
The flagship Mixture-of-Experts (MoE) instruction-tuned model released by Alibaba Cloud on July 21, 2025. The 2507 version represents a strategic pivot: it completely abandons the hybrid "thinking mode" in favor of a highly optimized Non-Thinking architecture. By removing the generation of <think> blocks, it provides significantly faster response times and improved relevance without sacrificing intelligence. It delivers state-of-the-art results across MMLU-Pro, AIME25, and coding benchmarks, outperforming leading competitors like GPT-4o and DeepSeek-V3 in non-reasoning efficiency. This model is the premier choice for enterprise RAG, high-precision instruction following, and complex multilingual tasks.
Prezzi
Performance
Utilizzo e classifica
Parametri supportati
Tutti i provider = supportato da ogni upstream che serve questo modello. Alcuni provider = dipende dall'upstream che gestisce la richiesta. Predefinito = il valore inviato quando non lo imposti.
| Parametro | Provider | Predefinito |
|---|---|---|
| include_reasoning | Tutti i provider | - |
| max_tokens | Tutti i provider | - |
| presence_penalty | Tutti i provider | - |
| reasoning | Tutti i provider | - |
| response_format | Tutti i provider | - |
| seed | Tutti i provider | - |
| temperature | Tutti i provider | - |
| tool_choice | Tutti i provider | - |
| tools | Tutti i provider | - |
| top_p | Tutti i provider | - |