# Qwen 4 vs Opus 5.5: Qwen4 Max costs less, Opus 5.5 scores higher

> Qwen4 Max output costs $4.50 per million tokens against $20 for Claude Opus 5.5, but Opus 5.5 leads Artificial Analysis's index 59 to 43.

URL: https://claudegpt.si/news/qwen-4-vs-opus-5-5  
Published: 2026-11-16T23:01:46.397Z  
Author: ClaudeGPT Si Newsroom  
Section: Reviews  
Topics: Qwen, Alibaba, Anthropic, Claude, Language models, API pricing  
Publisher: ClaudeGPT Si

## Key takeaways

- Qwen4 Max costs $1.50 per million input tokens and $4.50 per million output tokens, against $4 and $20 for Claude Opus 5.5, as of November 16, 2026.
- Artificial Analysis scores Opus 5.5 at 59 on its Intelligence Index at max effort and 48 at its default medium effort, against 43 for Qwen4 Max.
- Qwen4 Max accepts video input and longer outputs, while Opus 5.5 has a 50% Batch API discount that Qwen4 Max lacks.

In the Qwen 4 vs Opus 5.5 matchup, Alibaba's Qwen4 Max is the cheaper model and Anthropic's Claude Opus 5.5 is the stronger one. As of November 16, 2026, [Qwen4 Max costs $1.50 per million input tokens and $4.50 per million output tokens](https://www.qwencloud.com/models/qwen4-max), against [$4 and $20 for Opus 5.5](https://platform.claude.com/docs/en/about-claude/pricing). On Artificial Analysis, the independent leaderboard that runs the same tests on every model, Opus 5.5 scores 59 to Qwen4 Max's 43.

So the choice comes down to the job. For high-volume work where a cheap token matters more than the last few points of quality, Qwen4 Max is worth testing. For hard coding and long agent runs, the published numbers still favor Opus 5.5.

Alibaba [announced Qwen4 Max on November 16, 2026](https://www.alibabacloud.com/en/press-room/alibaba-unveils-qwen4-max). Anthropic [released Opus 5.5 on September 22, 2026](https://www.anthropic.com/claude-opus-5-5). For the launch details, prices by region and Alibaba's own benchmark chart, see our [Qwen4 Max release article](https://claudegpt.si/news/qwen4-max-api-price-release).

## Qwen 4 vs Opus 5.5 at a glance

| | Qwen4 Max | Claude Opus 5.5 |
| --- | --- | --- |
| Maker | Alibaba (Qwen team) | Anthropic |
| Released | November 16, 2026 | September 22, 2026 |
| API model ID | `qwen4-max` | `claude-opus-5-5` |
| Input price, per 1M tokens | $1.50 (Singapore), $1.413 (other regions) | $4 |
| Output price, per 1M tokens | $4.50 (Singapore), $4.24 (other regions) | $20 |
| Cached input, per 1M tokens | $0.075 | $0.20 |
| Batch discount | Not supported | 50% off ($2 in, $10 out) |
| Context window | 1M tokens | 1M tokens |
| Max output | 131,072 tokens without thinking, 262,144 with thinking | 128K tokens (up to 300K on the Batch API with a beta header) |
| Input types | Text, images, video | Text, images |
| Thinking | On by default at `high` effort | Always on, default effort `medium` |
| Artificial Analysis Intelligence Index | 43 | 59 at max effort, 48 at medium |
| Open weights | No | No |

Prices are from QwenCloud, Alibaba Cloud Model Studio and Anthropic's pricing page, as of November 16, 2026. This comparison is based on the companies' docs and published benchmark results, not on our own tests.

## How much Qwen4 Max and Opus 5.5 cost

Qwen4 Max is cheaper on every line of the price list. Input costs about 2.7 times less than Opus 5.5, and output about 4.4 times less. Output is usually the bigger part of the bill for a model that thinks before it answers, so the gap in practice sits closer to the second number.

[Alibaba Cloud Model Studio](https://www.alibabacloud.com/help/en/model-studio/qwen4-max) lists two prices. The Singapore region charges $1.50 and $4.50 per million tokens. Beijing, Hong Kong, Frankfurt, Virginia and Tokyo charge $1.413 and $4.24. The same price applies across the whole 1M-token context, with or without thinking. Singapore accounts also get 1 million free tokens, valid for 90 days.

Caching narrows the gap a little. Qwen4 Max charges $0.075 per million tokens for a cache hit, and $1.875 or $3 to write a 5-minute or 1-hour cache. Opus 5.5 charges $0.20 for a cache hit and $5 or $8 for the two cache writes. Anthropic says the full 1M-token window is billed at the standard rate.

Two things favor Anthropic on price. Its Batch API halves the price of Opus 5.5 to $2 in and $10 out for work that can wait, and Model Studio says Qwen4 Max does not support batch inference. Opus 5.5 also has a fast mode at $8 and $40 per million tokens, which Anthropic says runs up to 2.5 times faster. On the other side, Anthropic adds 10% to every price when a request is pinned to US-only inference.

A worked example: a request with 100,000 input tokens and 10,000 output tokens costs about $0.20 on Qwen4 Max in Singapore and $0.60 on Opus 5.5, before any caching.

## What the benchmarks show

Each company publishes its own scores, and they don't line up test for test. The independent numbers are the place to start.

[Artificial Analysis](https://artificialanalysis.ai/models/qwen4-max) gives Qwen4 Max 43 on its Intelligence Index, 38th of the 81 models it ranks, above the median of 41. [Claude Opus 5.5](https://artificialanalysis.ai/models/claude-opus-5-5) scores 59 at max effort, 5th of 81. Opus 5.5 still scores 48 at its default medium effort and 52 at high, both above Qwen4 Max. The pages list one Qwen4 Max entry and do not say which effort setting it ran at.

Artificial Analysis also measures what a task costs at list prices. One task of its index costs $3.27 on average with Qwen4 Max and $10.04 with Opus 5.5 at max effort. That is the cheaper model doing worse for about a third of the money, which is a fair summary of the whole comparison.

Speed goes the other way from price. Artificial Analysis measures Qwen4 Max at 53 output tokens per second through Alibaba's API and Opus 5.5 at 117 at max effort. Qwen4 Max took 139.46 seconds on average before its first answer token, against a median of 12.26 seconds for comparable models. Opus 5.5 took 14.14 seconds at medium effort and 368.87 seconds at max.

On the [Terminal-Bench leaderboard](https://www.tbench.ai/leaderboard), which tests agents working in a terminal, Opus 5.5 running in Claude Code at max effort shares first place on Terminal-Bench 4.1 with a 65.8% resolution rate. Qwen4 Max is not listed yet. Neither is it ranked yet on [Vals AI](https://www.vals.ai/benchmarks), which tests legal, finance and other industry work, where Opus 5.5 is second on the Vals RSI Index.

### What each company claims

| Claim | Source | Result |
| --- | --- | --- |
| Terminal-Bench 4.0, at xhigh effort | Anthropic | Opus 5.5: 66.4% |
| Humanity's Last Exam, with tools | Anthropic | Opus 5.5: 67.7% |
| OSWorld 2.1, computer use | Anthropic | Opus 5.5: 81.8% |
| Code Arena coding leaderboard | Alibaba | Qwen4 Max: 6th place |
| Work360, Alibaba's own industry test | Alibaba | Qwen4 Max beats Qwen3.8 Max at about 60% lower cost per task |

Anthropic's table compares Opus 5.5 with its own models and OpenAI's, not with Qwen. Alibaba's press release compares Qwen4 Max only with Qwen3.8 Max. Alibaba's launch post does put Opus 5.5 in a chart, and our [release article](https://claudegpt.si/news/qwen4-max-api-price-release) lists those scores. In that chart Qwen4 Max leads on long agent tasks and trails Opus 5.5 on terminal coding.

## Features and limits

Both models read up to 1 million tokens in one prompt, and both think before they answer. The differences are in the details.

- **Video input.** Qwen4 Max takes text, images and video. Opus 5.5 takes text and images.
- **Longer answers.** Qwen4 Max can write up to 262,144 tokens of thinking and answer combined. Opus 5.5 stops at 128K tokens on the regular API, or 300K on the Batch API.
- **Thinking controls.** Qwen4 Max offers `low`, `high` and `xhigh` effort, with `high` as the default. Opus 5.5 runs from `low` to `max`, with `medium` as the default, and its thinking can't be switched off.
- **Drop-in compatibility.** Model Studio serves Qwen4 Max through the OpenAI Chat Completions, OpenAI Responses and Anthropic Messages formats. Code written for Claude can try Qwen4 Max by changing the base URL and the model name.
- **Where to run it.** Opus 5.5 is on the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry. Qwen4 Max is on Alibaba Cloud Model Studio and QwenCloud, and [on OpenRouter](https://openrouter.ai/qwen/qwen4-max) at $1.425 in and $4.275 out.

Neither model has open weights. Alibaba says smaller Qwen4 models will be released openly "soon", without dates or sizes. Our article on [whether Qwen4 Max is open source](https://claudegpt.si/news/is-qwen4-max-open-source) has the details.

## Which one to use

Pick Qwen4 Max when cost per token decides the project: bulk document work, video input, or agents that run many cheap steps. Test it on your own tasks first, because the independent scores put it well below Opus 5.5 and it starts answering slowly.

Pick Opus 5.5 for hard coding, terminal work and long agent runs, where it leads Qwen4 Max on Artificial Analysis by 16 points and tops Terminal-Bench, and for work that can use the Batch API's 50% discount. Its default medium effort already scores above Qwen4 Max on Artificial Analysis, so you don't need max effort to get the gap.

Qwen4 Max only left preview on November 16, 2026. Expect Terminal-Bench, Vals and other leaderboards to add it over the coming weeks, and expect the comparison to change when they do.

## Frequently asked questions

### Is Qwen 4 better than Opus 5.5?

Not on the independent numbers so far. Artificial Analysis scores Qwen4 Max at 43 and Opus 5.5 at 59 at max effort, and Opus 5.5 at its default medium effort still scores 48. Qwen4 Max is much cheaper per token.

### Is Qwen 4 open source?

Qwen4 Max is not. Alibaba offers it only through its API and says smaller open-weight Qwen4 models will be released soon, without giving dates or sizes.

### Is Qwen 4 out?

Yes. Alibaba released Qwen4 Max on November 16, 2026, on QwenCloud and Alibaba Cloud Model Studio under the model ID qwen4-max.

### Can I use Qwen4 Max with code written for Claude?

Alibaba Cloud Model Studio serves Qwen4 Max through the Anthropic Messages format as well as OpenAI's formats, so Claude code can switch by changing the base URL and model name.

## Sources

- [Qwen4-Max model page](https://www.qwencloud.com/models/qwen4-max), QwenCloud
- [qwen4-max](https://www.alibabacloud.com/help/en/model-studio/qwen4-max), Alibaba Cloud Model Studio
- [Alibaba Unveils Qwen4-Max: New AI Teammate for Work, Creativity, and Discovery](https://www.alibabacloud.com/en/press-room/alibaba-unveils-qwen4-max), Alibaba Cloud
- [Pricing](https://platform.claude.com/docs/en/about-claude/pricing), Anthropic
- [Models overview](https://platform.claude.com/docs/en/about-claude/models/overview), Anthropic
- [Introducing Claude Opus 5.5](https://www.anthropic.com/claude-opus-5-5), Anthropic
- [Qwen4 Max: intelligence, performance and price analysis](https://artificialanalysis.ai/models/qwen4-max), Artificial Analysis
- [Claude Opus 5.5: intelligence, performance and price analysis](https://artificialanalysis.ai/models/claude-opus-5-5), Artificial Analysis
- [Terminal-Bench leaderboard](https://www.tbench.ai/leaderboard), Terminal-Bench
- [Vals AI benchmarks](https://www.vals.ai/benchmarks), Vals AI
- [Qwen4 Max on OpenRouter](https://openrouter.ai/qwen/qwen4-max), OpenRouter
