Qwen4-Max Arrives on Novita AI With 1M-Token Context

Qwen4-Max is now available on Novita AI as a serverless API, bringing native text, image, and video input, a displayed 977K-token context window, and up to 393,000 output tokens. Novita lists the model at $1.413 per 1M input tokens and $4.24 per 1M output tokens.
- Qwen4-Max is available on Novita AI as qwen/qwen4-max.
- It accepts text, image, and video input and displays a 977K-token context window.
- Novita lists a maximum output of 393,000 tokens in the API sample.
- Pricing is $1.413 per 1M input tokens and $4.24 per 1M output tokens.
- It is a separate Qwen4-family model from Qwen3.8-Max.
A new flagship with long-context and vision capabilities
Novita AI lists Qwen4-Max under the model ID qwen/qwen4-max as a new-generation flagship in the Qwen4 family. The company describes it as a model for long-horizon, production-grade work, including tasks that require several connected steps, planning, execution, and verification.
The hosted endpoint accepts text, image, and video input and produces text output. Novita lists a 977K-token context window, while the nominal context limit is 1,000,000 tokens. The model page displays a maximum output of 384K tokens, and its API sample uses 393,000 as the maximum output value.
According to Novita, Qwen4-Max is a 2.4T-parameter mixture-of-experts model with 71B activated parameters. The listing also includes reasoning, structured outputs, function calling, and serverless delivery.
How Qwen4-Max differs from Qwen3.8-Max
Qwen4-Max is a separate model from Qwen3.8-Max, rather than a checkpoint update or rename. The two models have different listings, model IDs, prices, and limits. Developers using qwen/qwen3.8-max will not be moved automatically to the new model, so switching requires a configuration change and a fresh evaluation.
The clearest capacity change is the output ceiling. Novita lists 393,000 maximum output tokens for Qwen4-Max, compared with 128,000 for Qwen3.8-Max, with 131,072 shown in the older model's API sample.
Novita also lists lower per-token prices for Qwen4-Max. Input costs $1.413 per 1M tokens and output costs $4.24 per 1M tokens, compared with $2 and $6 for Qwen3.8-Max. Cached input reads cost $0.071 per 1M tokens, compared with $0.25 for the previous flagship. Both current listings show text, image, and video input, so the main reasons to test the newer model are its generation quality, larger output capacity, and pricing profile.
API access and workloads to evaluate first
Qwen4-Max is available through Novita's serverless API, so users do not need to deploy the model. Novita provides an OpenAI-compatible base URL at https://api.novita.ai/openai and an Anthropic-compatible base URL at https://api.novita.ai/anthropic. The listed endpoint families include chat/completions and responses.
The model is aimed at workloads where the task, rather than a single turn, is the unit of work. Potential fits listed by Novita include long-horizon agents, multimodal document and video analysis, repository-scale coding, and large technical deliverables. A large context can hold source trees, issue history, and test output together, while the output limit can support a substantial report or code change in one response.
Novita advises testing the model rather than treating it as a default replacement. Teams should replay a fixed evaluation set, compare task success, tool-call validity, retries, latency, and total token cost, and retain the previous model as a rollback option. Short, high-volume requests, latency-sensitive interfaces, and text-only tasks that already work with smaller models may not benefit from Qwen4-Max's larger capacity. Image and video inputs can also affect token usage, while cached input reads may reduce the cost of repeated long prompts.


