Hardcore Reviews
Hardcore Reviews

Qwen4, ERNIE, ChatGPT: Three Calendars, Three Different Tenses

Line up three official calendars posted in November and the single label coming soon stops working: Alibaba Cloud spells out the Qwen4-Max-Preview window but gives no general-availability date, Baidu frames ERNIE 5.5 Flash Preview open-sourcing as a pledge not yet delivered, and OpenAI publishes a dated sunset schedule for legacy voice modes, December 11, 2026, with a fixed replacement path. One undated, one unfulfilled, one already scheduled. This comparison covers only what each official page says, what it leaves unsaid, and when a stated commitment lands: no benchmarks, no ranking, no recommendation, and no filling in blanks on behalf of any vendor. As of November 12, 2026, sources are limited to official product pages from each vendor, specification docs, and help-center notices.

Published November 13, 20269 min read
<!-- preview-state-products-review | review | Qwen4, ERNIE, ChatGPT: Three Calendars, Three Different Tenses -->

As of November 12, 2026, line up the official calendars three vendors posted during November and an awkward thing appears: they are routinely filed under a single word, "coming soon." Read each official page word by word, though, and the three are not in the same tense at all. Alibaba Cloud's Qwen4-Max-Preview spells out its preview window in detail but says nothing about when a general version arrives. Baidu's ERNIE 5.5 Flash Preview writes "open weights" as a promise that has not yet been kept. OpenAI, for the legacy ChatGPT voices, publishes a dated retirement schedule, down to which new entry point users will be moved to. One date undetermined, one action undone, one already scheduled: flattening all three into the same "coming soon" erases the difference that matters most.

This article reads only the public fields on those official calendars: what the official pages say verbatim, what they do not say, and the day on which a stated commitment should land. That is the only ruler used here. The sources are the vendors' own product pages, specification documents and help-center notices, with no second-hand summaries folded in and no benchmark scores, rankings or buying advice used to fill the gaps. Put plainly, this is a comparison of "said / not said / when it lands" across three official calendars, with zero scores, zero rankings and zero recommendations. Where an official page has put nothing on paper, the space is left empty rather than completed on the vendor's behalf.

That is also why this piece takes a deliberately different angle from several existing articles on the site. The three ERNIE 5.5 Flash Base pieces from November 6 (open weights, setup, and a benchmark comparison, see /en/posts/base-model-benchmark-comparison-review) cover the Base model's own specs and usage. The two Qwen4 architecture previews from August 30 and September 4 (see /en/posts/qwen3-8-flash-next-multimodal-hotspot) cover model structure and capability guesses. The two GPT-Live pieces from September 13 (see /en/posts/gpt-live-1-release-hotspot) cover the launch of the new voice model and its interface. Those go deep into a single subject; this article does not repeat single-product specs, does not walk through deployment, and does not compare scores. It sets three official statements side by side, comparing "said / not said / when it lands," not specs, deployment or benchmarks.

Three Statements, Three Tenses

One of the three is especially easy to misread. Seeing the word "preview," many people file it under "a new model is about to launch" and then invent a general-availability date from somewhere else. The official Alibaba Cloud pages, however, offer a different information structure.

Preview state: Qwen4-Max-Preview pins the window, leaves the general date blank

The banner on the Alibaba Cloud product page www.aliyun.com/product/qwen reads "Qwen4-Max preview edition debuts on Token Plan," but it is the specification document (updated November 11, 2026, 20:44:53) that draws the real boundaries. It gives the verbatim "Model ID: qwen4-max-preview" and states "currently a preview edition, supports Token Plan only, pay-as-you-go is not supported." The preview window is equally exact: "the preview period runs from 10:00 on November 11, 2026 to 24:00 on November 25, 2026 (UTC+8), with no changes during the period, and it will be taken offline when the preview ends." On specifications, the page lists "context length 1M Token" and "maximum output length 393,216 Token, with thinking and the final answer sharing this budget," plus "compatible with the OpenAI Chat Completions, Responses and Anthropic Messages APIs." Taken together, what is confirmed is a preview edition that is already callable, has a clear start and stop, and goes offline at the end: billing mode, context size, output cap and API compatibility are all on the page.

What matters is the part that is not written. Across the four official pages, parameter count, pay-as-you-go pricing, whether it is open source and under which license, and the general-availability release date are simply not mentioned. The accurate phrasing is one thing only: it is not mentioned. A blank is a blank; it is not a negative official statement, and it does not license anyone to supply a "will not" or a "has been delayed" on the vendor's behalf. One especially common slip is to claim the context size was not disclosed, which conflicts directly with the specification, since 1M is written out explicitly. The same document states that pay-as-you-go is not supported, so this article does not reverse-engineer a per-call price out of bundle and promotion wording either. The nature of this calendar can therefore be settled: a preview state, with a defined preview period and a general-availability date that is not mentioned. Because that date is not mentioned, "when will Qwen4 be released" has no answer on the official pages; only the preview window is written, and what comes after it is not.

Pledge state: ERNIE 5.5 Flash Preview writes open weights as a promise not yet kept

Baidu's Chinese release page (November 11, 2026) uses a different kind of wording. It states that ERNIE 5.5 Flash Preview is built on the previously released Base model, with 280B total parameters, 14B active parameters and 1M context, and it pledges verbatim that "the model API is live on the Qianfan platform, and the weights will be open sourced before November 25." The same page stops short of claiming too much, adding a self-qualification: "ERNIE 5.5 Flash Preview is an intermediate version in the post-training process ... and may still have some capability defects caused by insufficient training." A live API on one side and a self-described intermediate version on the other: the release page itself splits "already happened" from "still to come."

Read together, the nature is clear: the API being live is an accomplished fact, while open-sourcing the weights is a dated pledge, not something that has happened. Official wording names November 25 as the point in time but does not name a license; the weights have not been published, so there is no LICENSE text to rely on, and the accurate statement is "not announced, subject to the LICENSE text at the time." A supporting reference is the Base official page: the Base weights already released use a custom license, "ERNIE 5.5 Flash Base Model License Agreement v1.0," and no license name appears on the English pages or the product hub either. This article therefore neither writes the open-sourcing as completed nor rewrites the November 25 boundary into a release date. A single "will" pins this calendar to a pledge state: the words have been said, the event has not landed. What separates a pledge from an accomplished fact is exactly the step in which neither the weights nor the license text have appeared yet.

Sunset state: the legacy ChatGPT voices come with a fixed timetable

The third calendar is not about something arriving but about something leaving. OpenAI's retirement notice is titled "Retiring Advanced and Standard Voice in ChatGPT," states verbatim that "Advanced and Standard voice will be retiring on December 11, 2026." and lists in its timeline "November 10, 2026: Users are notified about the retirement." The transition is spelled out as well: "During the transition period: Users can continue using Advanced or Standard voice." "Users who previously selected these models will be defaulted to Live voice." It adds that "Conversations remain in your ChatGPT history and can be continued in text or voice." Even the exception is kept: "Live video and screen-sharing will continue to be powered by Advanced voice until GPT-Live has this capability."

This is a migration path with a date, an action and a fallback, and its nature is a fixed timetable, not a preview in any sense. Two wording boundaries are worth holding. First, the retirement notice uses "Advanced and Standard voice" throughout; the July 8 GPT-Live launch page uses a different phrasing for the same two legacy voices, the two pages describe different events, and this article keeps their wordings strictly apart rather than moving any figure from one page to the other. Second, what the notice writes is that the voices remain usable during the transition, that history is kept, and that the old options leave the model picker; it does not summarize a migration ratio, and it states dates and actions rather than an interval for readers to convert. Reading a fixed timetable as "coming soon" loses exactly the most concrete information: how it stops, how it switches, and what is preserved.

Field-by-Field

The table below moves only official fields, with no scoring and no recommendation. Keeping "said" and "not said / not announced" in separate columns is exactly what lets the blank, the pledge and the fixed timetable sit in their own cells.

ObjectNatureWhat was saidWhat was not said or announcedWhen it lands
Qwen4-Max-PreviewPreview stateModel ID and Token Plan billing, 1M and 393,216, two compatible APIs, preview from November 11 to November 25 and offline when it endsParameter count, pay-as-you-go pricing, open-source status and license, general-availability dateOffline at the end of the preview, general date not mentioned
ERNIE 5.5 Flash PreviewPledge state280B and 14B and 1M, Qianfan API live, weights to be open sourced before November 25, intermediate-version self-qualificationLicense not announced, subject to the LICENSE text at the timeBefore November 25, a pledge rather than an accomplished fact
Legacy ChatGPT voicesSunset stateRetirement on December 11, notification on November 10, usable during transition, defaulted to Live voice, history keptThe official page lists no arrangement beyond the replacement timetableDecember 11, 2026, a fixed timetable

The difference across the three rows is not how eventful they are but the verbs. The Qwen4 line says "taken offline when the preview ends," landing on the preview window itself. The ERNIE line says "will be open sourced before," landing on a pledge not yet kept. The legacy-voice line says "will be retiring on December 11, 2026" and "will be defaulted to Live voice," landing on a set of actions that will be executed. Calling all three "coming soon" compresses "not written," "not done" and "already scheduled" into one word. As long as the three columns stay separate, a pledge that is not an accomplished fact and a fixed retirement cannot be erased.

FAQ

Q1: When will Qwen4 be released? A1: It is not mentioned. As of November 12, 2026, the official Alibaba Cloud product page and specification document give only the preview period for qwen4-max-preview, from 10:00 on November 11, 2026 to 24:00 on November 25, 2026 (UTC+8), and state that it will be taken offline when the preview ends. As for a general-availability date, or whether there will be one, the official pages say nothing, so it can only be recorded as not mentioned rather than rewritten into a negative answer.

Q2: For ERNIE 5.5 Flash, is November 25 a launch or an open-sourcing? A2: It is the deadline of an open-source pledge, not a launch date. The Chinese release page states verbatim that "the model API is live on the Qianfan platform, and the weights will be open sourced before November 25": the API going live has happened, while open-sourcing the weights is a pledge to be delivered before November 25, a pledge rather than an accomplished fact. The official page names an open-sourcing deadline, not a launch day, and this article does not rewrite it into one.

Q3: Which license does ERNIE 5.5 Flash Preview use? A3: Not announced, subject to the LICENSE text at the time. The Preview weights have not been published; the vendor pledged open sourcing without naming a license, and no license name appears on the English pages or the product hub. The reference point is the already released Base weights, which use a custom license, "ERNIE 5.5 Flash Base Model License Agreement v1.0." Which license the Preview adopts can only be settled once the weights and the LICENSE text appear.

Q4: After December 11, 2026, can the legacy voices still be used? A4: Per the retirement notice, they can no longer be actively selected. "Advanced and Standard voice will be retiring on December 11, 2026." The notice says they remain usable during the transition, that the two options then leave the model picker, that users who previously selected them will be defaulted to Live voice, and that conversations remain in ChatGPT history and can be continued. "Live video and screen-sharing will continue to be powered by Advanced voice until GPT-Live has this capability" is the explicit exception. These are all part of a fixed timetable set by the vendor, not a preview still open for comment.

Q5: Can the three calendars be ranked against each other, with a recommendation? A5: No, and this article does not. The three sets of public fields differ in nature: one is a preview state with four items not mentioned, one is a pledge state with a license not announced, and one is a dated fixed timetable. There is no shared yardstick, so any ranking or recommendation would go beyond what the official pages say. All that can be offered is the "said / not said / when it lands" comparison and the reminder to read the three separately.

Related reading

For a deeper look at a single object, the site has more focused entry points: the open weights of ERNIE 5.5 Flash Base are covered in /en/posts/ernie-55-flash-base-open-weights-hotspot, at a different angle from the pledge state here, and the cost side of model retirement and migration is covered in /en/posts/model-sunset-migration-cost-sop. This article keeps its ruler on the public fields of the official calendars, and does not finish any sentence a vendor left unwritten.

This article is AI-assisted and human-edited. Last updated: 2026-11-13

FAQ

When will Qwen4 be released?
It is not mentioned.** As of November 12, 2026, the official Alibaba Cloud product page and specification document give only the preview period for qwen4-max-preview, from 10:00 on November 11, 2026 to 24:00 on November 25, 2026 (UTC+8), and state that it will be taken offline when the preview ends. As for a general-availability date, or whether there will be one, the official pages say nothing, so it can only be recorded as not mentioned rather than rewritten into a negative answer.
For ERNIE 5.5 Flash, is November 25 a launch or an open-sourcing?
It is the deadline of an open-source pledge, not a launch date.** The Chinese release page states verbatim that "the model API is live on the Qianfan platform, and the weights will be open sourced before November 25": the API going live has happened, while open-sourcing the weights is a pledge to be delivered before November 25, a pledge rather than an accomplished fact. The official page names an open-sourcing deadline, not a launch day, and this article does not rewrite it into one.
Which license does ERNIE 5.5 Flash Preview use?
Not announced, subject to the LICENSE text at the time.** The Preview weights have not been published; the vendor pledged open sourcing without naming a license, and no license name appears on the English pages or the product hub. The reference point is the already released Base weights, which use a custom license, "ERNIE 5.5 Flash Base Model License Agreement v1.0." Which license the Preview adopts can only be settled once the weights and the LICENSE text appear.
After December 11, 2026, can the legacy voices still be used?
Per the retirement notice, they can no longer be actively selected.** "Advanced and Standard voice will be retiring on December 11, 2026." The notice says they remain usable during the transition, that the two options then leave the model picker, that users who previously selected them will be defaulted to Live voice, and that conversations remain in ChatGPT history and can be continued. "Live video and screen-sharing will continue to be powered by Advanced voice until GPT-Live has this capability" is the explicit exception. These are all part of a fixed timetable set by the vendor, not a preview still open for comment.
Can the three calendars be ranked against each other, with a recommendation?
No, and this article does not.** The three sets of public fields differ in nature: one is a preview state with four items not mentioned, one is a pledge state with a license not announced, and one is a dated fixed timetable. There is no shared yardstick, so any ranking or recommendation would go beyond what the official pages say. All that can be offered is the "said / not said / when it lands" comparison and the reminder to read the three separately.

Related

Hardcore Reviews

TrnRT, Kindi, Baseten: One License Forbids Managed Service

Three inference engine repositories, TrnRT, Kindi and Baseten, on one comparison table, with license terms and hardware bets as the axis rather than a buying recommendation or benchmark ranking. Apache-2.0, Elastic License 2.0 and AGPL-3.0 coexist; one LICENSE says verbatim that the software may not be provided to third parties as a hosted or managed service, and GitHub reports NOASSERTION. Hardware claims from NKI megakernels on Trainium2 to GB300, B300, Quadro K420 and RTX 3080 are labelled as README self-reports. Stars are cumulative totals, with no growth score or single-day delta.

Nov 16, 20269 min read
Hardcore Reviews

qwen-code Preview, OpenBot, free-claude-code: Three Compared

Sets the qwen-code v1.0.0-preview beside OpenBot v1.0.2 and the rolling v6.x of free-claude-code in one table, comparing version states and access boundaries rather than picking a winner. Five rows cover models and providers, where AG-UI and MCP attach, permissions and approvals, stability during Preview, and license points. The free-claude-code column states AGPL-3.0-only, the NOASSERTION value from GitHub, the README disclaimer that it is not affiliated with Anthropic, and that provider terms govern integrations. Star counts are cumulative totals.

Nov 15, 20269 min read