AI Agent News/stories/alibaba-s-qwen4-max-now-available-on-together-ai-with-a-1m-token-context.md

Infrastructure

Alibaba's Qwen4 Max now available on Together AI with a 1M-token context

Together AI now serves Alibaba's flagship Qwen4 Max through its serverless inference, with a 1M-token context window. Sale pricing runs 30% off through November 22, 2026.

Source: www.together.ai/blog/qwen4-max

Together AI has added Qwen4 Max, Alibaba's flagship Qwen model, to Together Serverless Inference under the model string Qwen/Qwen4-Max. The model accepts text and image input, returns text, and supports tool calling. Prompt caching is automatic, and cached input is billed at a much lower rate than fresh input.

In its test, Together AI sent the full Tornado repository, 344 files and 652,519 tokens, together with a bug report in a single request and no retrieval step. The model returned a patch after 72 seconds, and the company says the patch passes the tests the Tornado maintainers wrote for that bug. The request cost about 69 cents at sale pricing.

Requests must be streamed, because the model accepts streaming only. Thinking is on by default and its tokens are billed as output, and it can be turned off with a reasoning parameter. Sale prices are $1.05 per 1M input tokens, $0.0525 per 1M cached input tokens and $3.15 per 1M output tokens, against list prices of $1.50, $0.075 and $4.50. The model is serverless only and cannot be fine-tuned or run on a dedicated endpoint.

Why it matters if you build agents

Agent builders can send a whole codebase in one call without writing their own retrieval step, but a request of that size takes time before the first token and costs far more without a cache hit. Repeated requests over the same content are much cheaper when the cache is hit, so prompt structure matters. Stream responses in code and budget for thinking tokens, which are billed as output.

Get stories like this every weekday

One short email with the day's agent tooling news.

Subscribe