Top line
Alibaba's Qwen4-Max is live in production at $1.50 input and $4.50 output per million tokens
Alibaba's Qwen team announced Qwen4-Max at 8:01 p.m. CT Sunday and said the API is live on QwenCloud. QwenCloud lists a 1M-token context window, text, image and video input, tool calling, structured outputs, and OpenAI- and Anthropic-compatible APIs. Pricing is $1.50 per million input tokens, $4.50 per million output tokens and $0.075 for implicitly cached input. Qwen says that is below Qwen3.8-Max; OpenRouter, which also listed the model Sunday night, gives the earlier price as $2 and $6. Qwen's claims of up to about 50% fewer tokens and a more than 60% lower cost per task come from its own internal evaluation, and we did not verify them.
The preview model qwen4-max-preview is separate. Alibaba's docs say it is available only through Token Plan, not pay-as-you-go, and will be retired after the preview ends at 24:00 UTC+8 on Nov. 25, which is 10:00 a.m. CT that day. The docs also say that in qwen4-max-preview, max_tokens now covers thinking plus the answer. Qwen also said open weights for smaller Qwen4 models will come, with no date given.
Why it matters: Teams on the preview need to switch to
qwen4-maxbefore Nov. 25 and re-size output limits. Anyone running Qwen3.8-Max can test the cheaper model, but should measure cost per task on their own workload.
Sources: Qwen announcement · QwenCloud model page · Qwen4-Max-Preview docs · Model Studio release list · OpenRouter pricing post
Tibo says Codex "model at capacity" errors have nearly disappeared; no reset or limit change
At 11:55 p.m. CT Sunday, Tibo Sottiaux wrote that the Codex team "might have actually fixed" the "model at capacity" errors. His attached chart shows the share of daily active users with at least one turn ending in a capacity error. It peaked at 6.82% in October and was 0.085% on Nov. 12. The chart marks six retry-related releases between Oct. 15 and Nov. 6. The data is OpenAI's own and the post hedges, so treat it as a strong signal rather than a guarantee.
This was not a usage announcement. We found no new reset, limit change or expiry date from Tibo. The banked reset he said was activated on Nov. 12 still has no stated plans, expiry or landing time.
Why it matters: If your team saw "model at capacity" failures in Codex last month, retry behavior should now be much better. Report any that persist.
Sources: Tibo's post · Tibo's Nov. 12 banked-reset post
What to do
- Move Qwen4-Max-Preview calls to
qwen4-maxbefore 10:00 a.m. CT on Nov. 25. Re-checkmax_tokens, because it now includes thinking. Docs - Monday, Nov. 16 deadlines. Federal agencies must fix the Check Point and UnrealIRCd KEV entries today. CISA Microsoft starts removing the remaining eMudhra roots from Windows, Edge, WebView2 and Azure Linux. Azure CLI 2.93.0 is due today and VS Code 1.147.0 on Nov. 18. MSRC
- Use Claude Fable resets on Pro and standard Team seats by 11:59 p.m. PT today (1:59 a.m. CT Tuesday). The week of included Fable use ends then too. Max, Ultra and premium-seat resets last until March 8, 2027, 11:59 p.m. PT. Anthropic
- GitHub's last pull_request_target brownout runs 10 a.m. to noon CT today. It affects public repositories only. Enforcement becomes permanent Dec. 2. GitHub
Also notable
- Anthropic: Claude errors for 73 minutes overnight. Elevated errors hit Sonnet 5.5 and Fable 5.2 across claude.ai, the API, Claude Code and Cowork from 1:32 a.m. to 2:45 a.m. CT. Status
- Tencent: Hy4 Flash priced at $0.07 input and $0.281 output per million tokens. Tencent Hy posted this at 12:38 a.m. CT, with $0.003 for cached input. It says the model matches Claude Haiku 5.5 in its own blind tests. We could not open the product page, so the claims are unverified. Tencent Hy post
- Fireworks: dataset creation, upload and download APIs now return 501. The change was dated Nov. 15. Listing, getting and deleting datasets still work, and training can read local or cloud files directly. Changelog
- Microsoft 365: preliminary post-incident report for last week's Exchange outage is out. Admins can read it in the admin center under MO1550821 and EX1550543. Microsoft 365 Status
- GitHub: pull request merge timeouts under investigation at 5:59 a.m. CT. The incident was still open at the cutoff. GitHub Status
Sources
Qwen announcement · QwenCloud model page · Qwen4-Max-Preview docs · Model Studio release list · OpenRouter pricing post · Tibo's post · Tibo's Nov. 12 banked-reset post · CISA · MSRC · Anthropic · GitHub · Claude status · Tencent Hy post · Fireworks changelog · Microsoft 365 Status · GitHub Status