Tasks and inputs
Qwen4 Max Preview is listed for Image-text-to-text, Text generation, Video-text-to-text. It accepts text, image, video and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 17, 2026
Qwen4 Max Preview was Alibaba's five-day early-access build of its Qwen4 flagship, offered only through a Token Plan subscription before GA qwen4-max.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: No public per-token price. Available only through Alibaba's Token Plan subscription (reported from $3/month for new users); trial reported to end 2026-11-25. Confirm on your own plan.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Five days. That is how long this preview stood on its own before Alibaba shipped the general-availability flagship on 16 November 2026. If you are reading this to decide what to build on, the short answer is: build on GA qwen4-max, and keep this page as a record of what the preview was and how to move off it.
The preview opened on 11 November 2026, and it was not on the normal pay-per-token menu. Alibaba Model Studio made qwen4-max-preview available only through Token Plan, its subscription product. Token Plan reportedly starts at $3 a month for new users. There was no public per-token API price for the preview; model listings show $0 per token on token-plan routes, which reflects the subscription billing rather than a free model. The trial window has been reported to end on 25 November 2026; check your plan's terms for the real date.
If you do not hold a Token Plan, you never had access to this model ID. Nothing is lost.
Alibaba's changelog gives it the same brief as the GA release: a multimodal flagship for text generation, image and video understanding and deep reasoning, aimed at coding, office productivity, front-end design and complex agentic work. Listed capabilities match too: web search, tool calling, structured outputs, and OpenAI- and Anthropic-compatible APIs.
One figure needs care. Some model listings give a maximum output of 393,216 tokens for the preview. Alibaba's own GA documentation gives 262,144 tokens in thinking mode and 131,072 without thinking, and does not document the preview's limit at all. We could not confirm the larger number on a vendor page. If you relied on outputs above 262K, test that on the GA model before migrating; do not assume it carries over.
The practical migration is a model-name change, but verify three things:
Save a dozen preview transcripts and replay them. Differences in tone or tool-call shape are easier to catch on known prompts than in production.
The preview sits in a long run of Max previews from Alibaba, including Qwen3.8 Max Preview and Qwen3.6 Max Preview. The GA model most teams ran before Qwen4 was Qwen3.8 Max. See all of them on the Alibaba provider page.
Compare the model on the work you need to complete.
Qwen4 Max Preview is listed for Image-text-to-text, Text generation, Video-text-to-text. It accepts text, image, video and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Documented limits: 1M context and 393.22K maximum output. Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Tencent
Tencent Hy4 Flash is the cost-focused companion to Hy4, keeping vision and agent skills at a far lower token price. Availability is still unsettled.
Alibaba
Qwen4 Max (qwen4-max) is Alibaba's first Qwen4-generation flagship: 1M-token context, text, image and video input, thinking on by default, $1.50/$4.50 per 1M.
Sarvam
Sarvam-105B-1711 is a refreshed open post-train of Sarvam's 105B MoE, tuned for tool calling, instruction following, agents and Indian languages.
IBM
Granite 4.1 8B LLMON Preview is IBM's model trained on block-structured LLMON prompts to resist indirect prompt injection in agents and RAG.
NVIDIA
Nemotron-Labs Sigma 8B is NVIDIA's open continuous diffusion language model, denoising 64-token blocks in parallel; English, GPU-only, 8K context.
speakleash
Bielik-11B-v3.1-Instruct is SpeakLeash's open Polish-first 11B instruct model, trained on ~56 languages, with five reasoning levels and tool calling.