Qwen3.8-Max is Alibaba's newest flagship model in the Qwen line and its highest-capability tier to date. Alibaba positions it directly against GPT-5.5, Claude Opus 4.7, and Gemini 3.1 Pro in its own migration guide, recommending it whenever a task needs the strongest available reasoning — complex multi-step analysis, deep logical derivation, and demanding agentic work. It is natively multimodal, listed by Alibaba under both text generation and image/video understanding: it accepts text, images, and video and returns text, with a 1M-token context window. The official capability matrix confirms full support for thinking mode, function calling, built-in tools, and structured outputs, making it a complete drop-in for agent frameworks and tool-calling pipelines. Qwen3.8-Max is served through three wire formats — OpenAI-compatible, Anthropic-compatible, and native DashScope — across Beijing, Singapore, Tokyo, Frankfurt, and Virginia. Thinking is toggled with the enable_thinking parameter (or reasoning.effort on the Responses API). It is the premium tier of the family: reach for it when correctness on hard problems outweighs cost, and drop to Qwen3.7-Plus or Qwen3.7-Flash for everyday volume.
Qwen3.8 Max is a multimodal large language model from the Qwen family, available through OrcaRouter with the model id 'qwen/qwen3.8-max'. The provider is qwen. It has a context window of 1,000,000…
Based on the catalog facts, Qwen3.8 Max is a multimodal language model that accepts text, image, and video inputs. That makes it suitable for tasks such as summarizing a long document, answering questions about an image, or analyzing video content. The model is expected to handle general text generation and reasoning, since it is part of the Qwen family of large language models. However, no specific task list or benchmark scores are included in the OrcaRouter listing. You should test the model with your own prompts to confirm it produces the style and accuracy you need. Because the API is OpenAI-compatible, you can send a small request through OrcaRouter without changing your existing code. The model's output tokens are billed at $6.00 per 1M tokens, so plan for generation costs as well as input costs. For best results, provide clear prompts and include only relevant context.
To use image and video inputs with Qwen3.8 Max, send them as content parts inside a chat message to OrcaRouter's OpenAI-compatible API. The standard OpenAI chat format allows a content array with a text part and an image_url part. Video input may require a specific format, which is not detailed in the catalog entry; the provider qwen lists video as an accepted modality. A common approach is to pass video frames as a sequence of images or use a provider-specific video encoding if OrcaRouter supports it. The model will then process the visual information together with the text prompt. Remember that all visual input is counted as tokens, and tokens are billed at $2.00 per 1M input tokens. If your video is long, the token count can be high, so test with a small sample first. Confirm the exact message schema in OrcaRouter's documentation before building production code.
Qwen3.8 Max is priced at $2.00 per 1M input tokens and $6.00 per 1M output tokens. If your prompts are short, your data is text-only, or you do not need a 1,000,000-token context, a cheaper model will likely give you similar results for a lower cost. Many text-only models on OrcaRouter have lower per-token rates and faster response times for tasks such as classification, extraction, summarization, and ordinary chat. You should choose Qwen3.8 Max when the task genuinely benefits from large context or from combining text with images or video. You should also compare the output quality on your specific workload, because price is not the only factor. For high-volume applications, a cheap model with acceptable quality is often the better business decision. Estimate the average input size per request and multiply by the rate to see where the break-even point falls.
The best use cases for Qwen3.8 Max follow from its listed capabilities: a 1,000,000-token context window and text, image, and video input. That includes long-document question answering, full-book summarization, contract analysis, codebase review, and multimodal tasks where you need to understand screenshots, charts, or video frames. It is also useful for processing an entire video transcript in one call, rather than splitting it into segments. Because output costs $6.00 per 1M tokens, you should aim for concise answers even when the input is long. If you only need to answer a short question from a small paragraph, this model is overkill and expensive. The model is a strong fit when the input is large, multimodal, or both, and the answer depends on all of that content. Use smaller models for high-volume generation tasks.
The catalog entry for Qwen3.8 Max on OrcaRouter does not list any benchmark scores. We are not supplying numbers from external evaluations because none are grounded in the provided facts. In general, benchmark scores measure a model's performance on tasks like general knowledge, reasoning, coding, and multimodal understanding, depending on the benchmark. Without official scores for Qwen3.8 Max, you cannot rely on a single metric. The appropriate way to assess the model is to run it on your own validation set using OrcaRouter's OpenAI-compatible API. Compare the outputs across several prompts and check consistency. If you later see benchmark scores published by the provider, treat them as one signal among many, not as proof your use case will work. A model that scores high on a broad benchmark can still fail on domain-specific input, so direct testing matters.
No latency or throughput figures are provided for Qwen3.8 Max in the catalog listing on OrcaRouter. Response speed depends on the qwen provider's infrastructure, the size of your input, and the current load on OrcaRouter. A request with 1,000,000 input tokens will take longer than a short prompt because the model must process the entire context before generating output. Output length also affects total time; generating many tokens takes time. If speed is critical, keep the input and output sizes as small as possible. You can measure time-to-first-token by streaming the response through OrcaRouter's API. Since no official speed figures exist, do not assume a particular response time. Run your own test with a realistic prompt size and measure it. Latency can also vary by region and time of day. The provider may throttle large requests.
The strengths of Qwen3.8 Max are grounded in the catalog entry: a context window of 1,000,000 tokens and support for text, image, and video inputs. This combination is useful for tasks that require reading a large amount of diverse content in one request. The honest limitations are that no benchmark scores or speed figures are listed, so you cannot verify quality from the catalog alone. The input price of $2.00 per 1M tokens is higher than many smaller models, and the output price of $6.00 per 1M tokens means long responses are not cheap. The model may not be the best choice for short, text-only, or latency-sensitive applications. You should evaluate Qwen3.8 Max on your own data through OrcaRouter before making it a production dependency. Rely on direct observation rather than marketing claims. For tasks that fit in a small window, a cheaper model is preferable.
Qwen3.8 Max is billed at the provider rate, which is $2.00 per 1M input tokens and $6.00 per 1M output tokens. OrcaRouter applies zero markup, so the token price you see is the token price you pay. There is no mention of a fixed per-request fee in the catalog entry. The cost of a request is calculated by counting every input token, including text, image, and video tokens, and multiplying by $2.00 per 1M tokens. Output tokens are counted separately and multiplied by $6.00 per 1M tokens. For example, a request with 250,000 input tokens and 5,000 output tokens costs $0.50 for input and $0.03 for output. Actual charges will appear on your OrcaRouter invoice. If you use the full 1M context, the input cost alone is $2.00. There are no other listed fees.
The full context window of Qwen3.8 Max is 1,000,000 tokens. At $2.00 per 1M input tokens, a request that uses the entire window costs $2.00 for input before any output is generated. If the output is substantial, you also pay $6.00 per 1M output tokens. Visual inputs consume tokens quickly, so including video frames or many images can raise the input token count far above what you might expect from text alone. This pricing model means there is no fixed cost per call; you only pay for the tokens used. It also means that a prompt with 100,000 tokens costs $0.20, while a prompt with 1,000,000 tokens costs $2.00. If your task only needs a few thousand tokens of context, the value of Qwen3.8 Max is harder to justify. Consider smaller models for those cases.
The catalog entry for Qwen3.8 Max does not mention caching. OrcaRouter's OpenAI-compatible API may support standard provider-reported cache hits, but the model facts do not confirm that. Without confirmed caching, you should assume that every input token is billed at the standard rate of $2.00 per 1M tokens. Some providers automatically cache a prefix of your prompt and charge less for repeated tokens, but that is not stated for this model. You can check the usage object in OrcaRouter's API response for cached_token fields; if the provider reports them, you will see the discount. If not, budget for full input cost per request. The safest plan is to test with repeated identical prompts and inspect the token usage in the response. Caching, if absent, will not reduce your bill.
To call Qwen3.8 Max, use OrcaRouter's OpenAI-compatible API at base URL https://api.orcarouter.ai/v1. Set the model id to 'qwen/qwen3.8-max' in the request body. The endpoint is https://api.orcarouter.ai/v1/chat/completions. You send a JSON object with a messages array, where each message has a role and content. For text-only input, content is a plain string. For image or video input, content can be an array of parts, following the OpenAI vision format. Authenticate with your OrcaRouter API key in the Authorization header. OrcaRouter routes the request to the qwen provider. No separate provider-specific base URL is needed. Use a standard OpenAI SDK and set base_url to OrcaRouter's URL to get started quickly. The response follows the same chat completions format you already know.
Through OrcaRouter's OpenAI-compatible API, you can set parameters such as temperature, top_p, max_tokens, and stop sequences. The supported parameters may depend on the qwen provider's backend. Qwen3.8 Max has a context window of 1,000,000 tokens, so the combined input and output tokens must stay within that limit. The maximum output length is not listed in the catalog entry; check the model documentation or error messages for that limit. You can use the stream parameter to receive tokens as they are generated, which can improve perceived latency. For multimodal inputs, the message content array needs to include the correct part types. If a parameter is not supported, OrcaRouter will return an error, and you can adjust your request. Test with a small payload first to confirm parameter behavior. This is standard practice when integrating any new model.
If your application already uses OpenAI's chat completions API, migration to Qwen3.8 Max on OrcaRouter is straightforward. Change the base URL to https://api.orcarouter.ai/v1 and set the API key to your OrcaRouter key. Set the model field to 'qwen/qwen3.8-max'. The message structure stays the same, including system, user, and assistant messages. For multimodal requests, use the same content parts format as OpenAI's vision API. You may need to update your prompts because Qwen3.8 Max is a different model and may respond differently. Test with a few sample requests before changing production traffic. Also note that the model id includes the 'qwen/' prefix, which is how OrcaRouter identifies the provider. No code rewrite is necessary if your SDK allows a custom base URL. This reduces migration effort. You can keep the same retry and error handling logic.
Qwen3.8 Max is distinguished by its 1,000,000-token context window and its support for text, image, and video inputs. Smaller Qwen models usually have shorter context windows and may not accept all three modalities. Because no benchmark scores are provided for Qwen3.8 Max on OrcaRouter, a direct quality comparison against smaller models is not possible from the catalog facts. The pricing difference is clear: Qwen3.8 Max charges $2.00 per 1M input tokens and $6.00 per 1M output tokens. Smaller models typically charge less per token and can be faster, but their limits may force you to chunk inputs or drop visual data. If your project fits within a smaller context and does not need video input, a smaller model is likely more economical. If you need to reason across a long document or video, Qwen3.8 Max is the option designed for that.
OrcaRouter hosts multiple models from different providers, and Qwen3.8 Max is one of the multimodal options. Its defining features are a 1,000,000-token context window and text, image, and video input capabilities. Other multimodal models may have smaller context windows or different token pricing. The catalog entry for Qwen3.8 Max lists $2.00 per 1M input tokens and $6.00 per 1M output tokens, with zero markup on OrcaRouter. Without benchmark scores, you cannot conclude which model is more capable. You can compare them directly by sending the same prompts to each model via OrcaRouter's OpenAI-compatible API and comparing output quality, response time, and cost. The choice depends on your specific workload and how much context you actually need. Video support is not universal, so that is a key differentiator. A model with lower price may still be better if your prompts are small.
Pick Qwen3.8 Max when the task requires a large context window of up to 1,000,000 tokens or demands multimodal input with text, image, and video. It is a reasonable choice for long-document analysis, whole-video understanding, and image-plus-text reasoning. Pick an alternative when your prompts are short, text-only, or latency-sensitive, and when a smaller model with a lower token price can deliver acceptable quality. Also consider alternatives if you need published benchmark scores or guaranteed throughput, since those are not listed for Qwen3.8 Max. The right decision depends on your expected token usage, cost tolerance, and quality requirements. Run a small evaluation on OrcaRouter with both models and calculate total cost per successful answer before committing to production. There is no single best model for every task.
OpenAI-compatible — keep the SDK you already use
https://api.orcarouter.ai/v1https://api.orcarouter.aiimport os
from openai import OpenAI
client = OpenAI(
base_url="https://api.orcarouter.ai/v1",
api_key=os.environ["ORCAROUTER_API_KEY"],
)
response = client.chat.completions.create(
model="qwen/qwen3.8-max",
messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)enable_thinkinginclude_reasoningmax_tokenspresence_penaltyreasoningreasoning_effortresponse_formatseedstopstructured_outputstemperaturetool_choicetoolstop_p| Input / 1M tokens | $2.00 |
| Output / 1M tokens | $6.00 |
| Cache read / 1M | $0.250 |
| Cache write / 1M | $2.50 |
| Currency | USD |
Estimate based on list price
Estimate only — actual token counts depend on the provider's tokenizer.
What developers are saying this week
GET /api/public/models/qwen/qwen3.8-maxOpen @misc{orcarouter_qwen3_8_max,
title = {Qwen3.8 Max API},
author = {Qwen},
year = {2026},
howpublished = {OrcaRouter},
url = {https://www.orcarouter.ai/models/qwen/qwen3.8-max}
}Qwen. (2026). Qwen3.8 Max API. OrcaRouter. https://www.orcarouter.ai/models/qwen/qwen3.8-max