Qwen3.8-Flash-Next is live on QwenCloud at $0.16 per million input tokens and $0.47 output, against $2.00 and $6.00 for Qwen3.8-Max, roughly a 12x gap on both, per The Decoder. In Alibaba's published results it scored 62.5 on SWE-bench Pro to DeepSeek-V4-Flash's 56.0, with a native 262K context and a permissive commercial license on Hugging Face. Price your agent loop against this number before you renew a frontier contract for work a 6B-active MoE can do. Read the benchmark detail.

Business Standard and Plus, Enterprise Standard and Plus, and Google AI Pro and Ultra all get it; Rapid Release domains started September 1 and Scheduled Release domains land September 15, per the Workspace Updates blog. One feature justifies the trial: hover-select the text inside an image and translate it while the design and font survive. Marketing teams about to buy a Canva or Adobe Express seat for exactly that localization job should run the comparison this week. See what ships and when.

Read-only pulls of appointment notes, labs, medications, and specialist docs, plus a plugin spanning nine public sources including ClinicalTrials.gov, RxNorm, and PubMed, per OpenAI. The EHR channel requires an enterprise account and a signed BAA; the public-data plugin does not, so the rollout order writes itself. Give clinicians the plugin now while legal negotiates the BAA. OpenAI's physician panel rated 99.1% of responses safe across 4,363 ratings, which is a vendor evaluation and a floor to verify, not an audit. Read the governance model.

Critical, under OpenAI's Preparedness Framework, means finding and chaining zero-day exploits in hardened production systems with no human in the loop. Astra weaponized two zero-days in a single exploit chain during evaluation, and OpenAI paused some internal work in August to add controls, per TechCrunch. Full cyber capability goes only to vetted defenders in the Daybreak Blue program; general API users get a filtered model. If you run a red team or sell security tooling, check eligibility now, not the week the API opens. Read what Critical actually means.

OpenAI pulls its models out of Cursor on November 12, and our rating drops to 4.2 from 4.4. The call splits by team size. Solo devs already defaulting to Claude or Gemini feel nothing, so stay and stay monthly. Teams of 3 to 10 should refuse an annual contract until Cursor publishes a written post-split model-access commitment. Anyone renewing 50-plus seats needs that guarantee, plus data routing, in the contract itself. Read the November 12 call or go straight to Cursor.