Model by Alibaba (Qwen)
Qwen3.8 2.4T A95B is the open-weight version of Alibaba's Qwen3.8 Max, for teams with a serious GPU cluster who want a frontier-size model they can host.
Checked 2026-11-05
Loading poll...
Qwen3.8 2.4T A95B is the open-weight version of Alibaba's Qwen3.8 Max, for teams with a serious GPU cluster who want a frontier-size model they can host. The Hugging Face card gives 2.4 trillion parameters in total with 95 billion active per token, a native context of 262,144 tokens that can stretch to 1,010,000, and an August 2026 date. It takes text only, thinking cannot be switched off, and every reply opens with a reasoning block. The card recommends recent SGLang, vLLM or TokenSpeed builds and states no memory needs. The licence is a custom Qwen3.8-Max licence, not Apache. Use is free, but a company running a model-as-a-service or coding-assistant business over $50 million in twelve months needs a separate licence, and large products must show the model name. OpenRouter lists it from 12 August 2026 at $2 per million input tokens and $6 per million output tokens on its cheapest route as listed on 5 November 2026. We have not run it ourselves. We would use the API route first and read the licence before self-hosting.
Qwen3.8 Max (0902) is a closed Alibaba model released 2026-09-03, listed on OpenRouter at $2 per million input tokens and $6 output with a 1,000,000-token context and text, image and video input.
Alibaba (Qwen), 1M context, $2 in / 1M
Qwen3.8 Max Prime is a higher-throughput version of Alibaba's Qwen3.8 Max, sold as its own SKU at a higher price.
Alibaba (Qwen), 1M context, $4 in / 1M
Qwen3.8 27B is an open model from Alibaba's Qwen team, released 2026-08-14 under Apache 2.0, a standard permissive licence.
Alibaba (Qwen), 1M context, $0.15 in / 1M
OpenRouter routes more than 600 models through one API key.
vLLM is a high-throughput, memory-efficient serving engine for large language models, run as a server you host yourself.
SGLang is a serving framework for large language and multimodal models, aimed at low latency and high throughput from a single GPU up to large clusters.
Checked 2026-10-31
Checked 2026-10-31
Checked 2026-10-31
1,048,576 tokens, with up to 131,072 tokens of output.
$2 per million input tokens and $6 per million output tokens, checked 31 October 2026.
Yes, the weights are published under the Open-weights model licence. They are on huggingface.co.
Alibaba (Qwen) makes it, and released it on 12 August 2026.
Loading poll...
Loading reviews…
Only members can review this page. Sign in
Qwen3.7 Max is Alibaba's closed flagship text model from the Qwen3.7 series, aimed at agent, coding and office work for people who will pay for a hosted API rather than run weights.
Alibaba (Qwen), 1M context, $1.48 in / 1M
Qwen3.7 Plus is Alibaba's hosted multimodal model, and it suits teams that want image and video understanding plus agent tool use through an API rather than downloadable weights.
Alibaba (Qwen), 1M context, $0.32 in / 1M
Qwen3 Coder Next is an open-weight coding model from Alibaba's Qwen team, and it suits people who want a capable agent model they can run on their own hardware.
Alibaba (Qwen), 262K context, $0.12 in / 1M
Qwen3.8 Omni Flash is Alibaba's omni-modal reasoning model, described by OpenRouter as the first Qwen model built around agentic capabilities with native audio and video understanding.
Alibaba (Qwen), 1M context, $0.15 in / 1M