Qwen 3.8 Max Preview – What the Community Knows

Qwen 3.8 Max Preview – What the Community Knows

Qwen 3.8 Max Preview is a subscription‑only model

The Qwen 3.8 Max Preview model is listed on the Qwen Cloud pricing page under a token‑plan subscription, not on a dedicated announcement page. The banner on the Qwen website redirects to the pricing page, indicating that the model is currently offered only to paying customers.

Token‑plan limits and configuration

A user who succeeded in configuring the model for OpenCode shared the exact JSON snippet required:

"provider": {
  "alibaba-token-plan": {
    "models": {
      "qwen3.8-max-preview": {
        "limit": { "context": 1048576, "output": 65536 },
        "modalities": { "input": ["text"], "output": ["text"] },
        "name": "Qwen3.8 Max Preview"
      }
    }
  }
}

The limits (1 MiB context, 64 KiB output) match those of the earlier Qwen 3.7 Max model, suggesting that the new preview inherits the same token caps.

Community impressions of Qwen 3.8 Max

Mixed performance expectations

  • Verbosity concerns: One commenter recalled that Qwen 3.7 Plus and Max became overly verbose, producing multi‑paragraph answers where concise replies were expected. They hope Qwen 3.8 Max will be less wordy.
  • Benchmark skepticism: The same user expressed distrust in the published benchmark scores for Qwen models and asked how Qwen 3.8 Max compares to competitors such as Sonnet 5, GLM‑5, DeepSeek V4 Pro, Gemini Flash, or Gemini 3.5 Flash.
  • Negative experience with Qwen 3.7 Pro: Another user reported that after a month of daily use, Qwen 3.7 Pro was “unusable” due to off‑track responses, endless loops, and high cost relative to DeepSeek V4 Pro. They described it as the worst software‑engineering experience among large language models.
  • Positive API feedback: A different commenter, who has used Qwen models since version 2.5, praised the API for being “quite good” and cheaper than OpenAI while delivering comparable quality. They intend to try the new preview.

Pricing and token economics remain opaque

A question about the meaning of “credits” and their conversion to tokens for each model went unanswered, indicating that the token‑plan pricing model lacks clear public documentation.

Open‑source availability and future plans

  • Open model uncertainty: Users asked whether Qwen 3.8 will ever be released as open‑source weights, noting that Qwen 3.6 was the last model with publicly available weights.
  • Potential future release: One comment referenced an X (formerly Twitter) post suggesting that Alibaba may eventually release the model openly, though no concrete timeline was provided.

Operational and legal considerations

  • Privacy policy request: A community member sought the token‑plan privacy policy to determine if user inputs or outputs are retained or used for training, but no official policy link was shared.
  • Ownership clarification: Another user asked who operates the pricing site, suspecting it could be an Alibaba frontend or a Singapore‑based reseller; no definitive answer was given.

Summary of the current state

  • Qwen 3.8 Max Preview is accessible only via a paid token plan, with limits identical to Qwen 3.7 Max.
  • Early user feedback is split: some praise the API’s cost‑effectiveness, while others criticize verbosity, reliability, and cost compared to alternatives like DeepSeek V4 Pro.
  • Critical details—pricing conversion, privacy policy, and open‑source release plans—remain undocumented, leaving potential adopters with unanswered questions.

Sources