Claude Opus 5.2 has a release date in exactly the sense that one X account has posted one. At 07:57 UTC on September 14, 2026, @notjazii listed it in a launch-week wish list next to Gemini 4, Grok 4.7 and the GPT-6 family. At 12:15 UTC the same account wrote that “opus 5 is getting routed to opus 5.2” inside Claude Code, because one prompt on the highest effort setting had come back cleaner there than in Devin. At 14:00 UTC that became “opus 5 is confirmed routing opus 5.2”, with a prompt others could run; at 14:38 UTC, “mostly coming this thursday or friday”. Anthropic has published nothing: no post, no identifier, no card, no statement.
As of 16:10 UTC on September 14 we read Anthropic’s model documentation, its platform release notes, its news index, its status page and the Anthropic section of OpenRouter’s catalog. None contains the string 5.2. The documentation lists four current models, Claude Fable 5.1, Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5, and the newest Opus identifier is claude-opus-5, released July 24. Our Opus 5.1 tracker carries the August version of this story, which ran on early-access strings that turned out to belong to Fable 5.1.
So this page does what our Fable 5.2, Grok 4.7 and Opus 5.1 trackers do: every claim dated, the single source separated from the accounts repeating it, the company’s own surfaces checked and named, and the page rewritten on this URL the day Anthropic moves. The wrinkle this time is the evidence itself. There is no leaked string and no screenshot of a model list. There is a prompt, and the claim that the answer to it tells you which model you are talking to.
On this page · 12 sectionsOpen
Claude Opus 5.2 does not exist in any Anthropic document as of September 14, 2026. The model documentation lists Claude Fable 5.1, Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5; the release notes stop at September 10; the news index, the status page and OpenRouter’s Anthropic catalog carry no 5.2.
Every claim traces to one X account, @notjazii, on September 14: at 12:15 UTC that Opus 5 is getting routed to Opus 5.2 inside Claude Code, based on the same prompt answering differently in Claude Code and Devin; at 14:00 UTC that the routing is confirmed; at 14:38 UTC that the release is mostly coming this Thursday or Friday.
The evidence offered is a prompt, not a document: ask Claude Code whether it knows who tibo, the reset guy, is, with search off. A yes is read as Opus 5.2. What the prompt actually measures is recent knowledge, which a newer checkpoint would show but so would retrieval or a system prompt. Anthropic dates Opus 5’s training data to May 2026.
Silent routing to a new model is a real Anthropic practice: the Opus 5.1 rumor in August ran on early-access strings and reported A/B traffic. This time there is no string, no identifier and no traffic report, only answers that feel different. A quiet swap and a quiet prompt change look identical from the outside.
The dates already disagree. Sunday: end of this month or at most end of next. Monday morning: this Thursday or Friday. Monday afternoon, from a second account: at the end of this month. None of them comes from Anthropic, which does not pre-announce models.
Polymarket prices the next Claude Opus at 80% by September 30, up from a market that resolved no on August 31, and above 97% by year end. That is a read on how many people believe the posts, not evidence of a model.
This is a living tracker. An anthropic.com/news entry, a claude-opus-5-2 identifier in the docs, a system card, a provider catalog line or a dated Anthropic statement flips it into a record with a scorecard of every claim above, on this URL.
§ 01What we actually know
| Claim | Status |
|---|---|
| Opus 5 in Claude Code is being routed to a new model | @notjazii on X, 12:15 UTC September 14, from one prompt answering differently in Claude Code and Devin; no Anthropic document |
| The new model is Opus 5.2 | The same account’s label; no identifier, no string, no screenshot carries the name |
| “opus 5 is confirmed routing opus 5.2” | The same account, 14:00 UTC; the confirmation is more tests by the same user, plus a prompt others can run |
| The routed model is faster, cleaner, not lazy | The same account’s description of its own sessions |
| Release “this thursday or friday” (September 17 or 18) | @notjazii, 14:38 UTC; no Anthropic date |
| Release by the end of September | @goodworse, 15:47 UTC, citing Polymarket; @pradeepXkapoor on September 13, “by max end of next month” |
| A test prompt identifies the model | The tibo prompt tests recent knowledge, not a version; see below |
| Leaked benchmark sheet, model identifier, early-access string | None exists |
| Anthropic’s model docs, release notes, news, status page, OpenRouter | No Opus 5.2; four current models, newest Opus is claude-opus-5 (read 16:10 UTC September 14) |
| Anthropic’s newest Opus | Claude Opus 5, July 24, 2026: 1M context, 128K output, $5 / $25 per MTok, training data May 2026 |
§ 02What the sources said, word for word
Every post below was read directly on X on September 14 and its time is taken from the post itself, so the tracker does not depend on the write-ups that followed.
@notjazii, 07:57 UTC: “Happy Launch Week”, with a list: “gemini 4”, “grok 4.7”, “opus 5.2”, “gpt 6 family”, “few chinese models”. No claim about any of them yet; the name is on the wish list.
@notjazii, 12:15 UTC: “opus 5 is getting routed to opus 5.2”, then the method: “tested opus 5 on xhigh with the same prompt in devin and claude code” and “results came out really different”, the Claude Code answer “clean and more detailed”, the Devin one “almost there but not the same”. Also: “opus 5 in cc used to be lazy”.
@notjazii, 14:00 UTC: “opus 5 is confirmed routing opus 5.2”, after more tests of its own: the routed mode is “way faster”, “gives really clean output”, “not lazy and loves to do longer tasks”. Then the prompt to run in Claude Code: whether the model knows who tibo, the reset guy, is, with search off; if it does, you have the new model. Asked in the replies “In what way is this a confirmation?”, the account did not add a document.
@notjazii, 14:38 UTC: “and opus 5 is getting routed to opus 5.2 so it’s mostly coming this thursday or friday too”, in a post that also gives “high chances for grok tomorrow if not then thursday”. At 15:53 UTC: “opus 5.2 really high chances but not sol”.
@goodworse, 15:47 UTC: “Opus 5.2 is coming AT THE END of this MONTH”, “Polymarket gives an 81% chance of a September release”, “the model is being tested as Opus 5 in Claude Code”, crediting @notjazii for the information five minutes later.
@pradeepXkapoor, September 13, 09:05 UTC, a day before any routing claim: “Opus 5.2 coming soon! By the end of this month most probably or by max end of next month”. No source given.
The name is older than the rumor. On March 20, 2026 the researcher Tim Hua wrote, labelled “Prediction (based on no private information)”, that “Before Claude Opus 5.2, Anthropic would publicly announce the existence & benchmarks of an internal model specialized in AI R&D”. That is the earliest use of the name we can find, and it treats 5.2 as the obvious next number, not as news.
§ 03The chain, dated
| Date (2026) | Event | Source |
|---|---|---|
| Jul 24 | Claude Opus 5 released (claude-opus-5) | Anthropic |
| Aug 21 to 30 | Opus 5.1 rumor: early-access strings, reported test traffic; no Opus 5.1 ever ships | Community, see our Opus 5.1 tracker |
| Sep 1 | Claude Fable 5.1 and Mythos 5.1 released; the August strings resolve toward Fable | Anthropic |
| Sep 13, 09:05 UTC | “Opus 5.2 coming soon! By the end of this month most probably” | @pradeepXkapoor on X, no source |
| Sep 14, 07:57 UTC | Opus 5.2 on a launch-week wish list | @notjazii on X |
| Sep 14, 11:19 UTC | Musk grades Grok 4.7 “on par with Opus 5.0, not 5.1”; no 5.2 | @elonmusk on X |
| Sep 14, 12:15 UTC | “opus 5 is getting routed to opus 5.2” in Claude Code, one prompt vs Devin | @notjazii on X |
| Sep 14, 14:00 UTC | “confirmed routing”, the tibo test prompt | @notjazii on X |
| Sep 14, 14:38 UTC | “mostly coming this thursday or friday” | @notjazii on X |
| Sep 14, 14:23 to 15:52 UTC | A dozen accounts repeat the routing claim; Polymarket by Sep 30 read at 80% | X, Polymarket |
| Sep 14, 16:10 UTC | Anthropic docs, release notes, news, status, OpenRouter: no Opus 5.2 | Our read |
The rumor is nine hours old at the time of writing and already carries two dates. What has not moved is the evidence: still one account’s sessions, still no artifact.
§ 04The test prompt, and what it can and cannot show
The rumor’s one checkable claim is the prompt. Ask Claude Code whether it knows who tibo, the reset guy, is, with search off; if it does, the posts say, you are talking to Opus 5.2. It is worth being exact about what a yes proves.
A model that knows a recent internet fact has that fact from somewhere: training data that extends past the fact, a system prompt or memory that mentions it, or a tool that fetched it. Anthropic’s model page dates Claude Opus 5’s training data to May 2026, so a confident answer about a later meme is a genuine signal that something has more recent knowledge than the documented cutoff. It does not say what. A new checkpoint of Opus 5 would pass the test. So would a new model under a new name, a prompt or memory change in Claude Code, or a different effort setting producing a different answer to an ambiguous question. The two accounts that ran the test both report a yes and both label it Opus 5.2; the label is theirs.
This is the difference from August. The Opus 5.1 rumor rested on two early-access identifiers seen in third-party model lists and a reported traffic pattern with a fallback to Opus 4.8: artifacts a reader could look for. This one rests on the feel of the answers and one knowledge probe. A quiet model swap and a quiet prompt change look the same from the outside, and only Anthropic can say which, if either, happened.
§ 05What Anthropic’s own rhythm says
Anthropic does not pre-announce models, so the only calibration is its shipping record. Here is 2026 to date, from Anthropic’s own news pages, with the rumored date applied.
| Model | Released | Days since previous |
|---|---|---|
| Claude Fable 5 (with Mythos 5) | June 9, 2026 | first of the line |
| Claude Sonnet 5 | June 30, 2026 | 21 |
| Claude Opus 5 | July 24, 2026 | 24 |
| Claude Fable 5.1 (with Mythos 5.1) | September 1, 2026 | 39 |
| Opus 5.2, if September 18 as rumored | rumored | 17 (56 after Opus 5) |
| Opus 5.2, if September 30 as rumored | rumored | 29 (68 after Opus 5) |
Two readings. Across the line, Anthropic shipped something roughly monthly this summer, and 52 days have passed since Opus 5, so an Opus point release in the coming weeks is inside its pace. Against the last release, a model 17 days after Fable 5.1 would be the fastest turn of Anthropic’s year, and that is the date the source gives. The end-of-month dates from the other accounts sit closer to the pattern. Either way, the cadence argues that a new Opus is plausible and says nothing about whether one is already being served.
§ 06What the market says
| Question | Price | Volume |
|---|---|---|
| By August 31, 2026 | resolved no | $41.5k |
| By September 30, 2026 | 80% | $48.9k |
| By October 31, 2026 | 98% | $13.3k |
| By December 31, 2026 | 98% | $5.1k |
The market opened on July 28, four days after Opus 5 shipped, and its August leg has already resolved no. The September leg moved on the day’s posts; @goodworse quotes it as the reason to expect a release. It is a read on how many people believe @notjazii, which is a different question from whether Anthropic has a model.
§ 07What would confirm it
Five things, any one of which flips this page from tracker to record:
- An entry on anthropic.com/news, where Opus 5, Sonnet 5 and Fable 5.1 all appeared on their release days.
- A new identifier in Anthropic’s model documentation, beside claude-opus-5.
- A system card at anthropic.com/system-cards.
- A catalog line on a provider: Amazon Bedrock, Google Cloud, Microsoft Foundry or OpenRouter.
- A dated statement from Anthropic naming the model.
What would not confirm it: a screenshot of a Claude Code answer, a prompt that gets a knowledgeable reply, a second account repeating the first, or a prediction market moving on the posts. Those go in the rumor table.
§ 08Where CellCog stands
Our conflict, declared: we build CellCog, a platform where a business hires AI employees that keep memory and work as a team, and we choose the models underneath them. Our Creative tier runs on Claude Opus 5; our Core and Max tiers run on Claude Fable 5.1, moved there on its release day. If Anthropic is serving a newer model behind the Opus 5 identifier, we would see it in the same place everyone else does, in the answers, and we would not know either. An announced Opus 5.2 becomes a same-day routing decision for the Creative tier, made on its price, its context window and its behavior on long creative runs. This page flips on the same URL when that day comes, with a scorecard of every claim above.
§ 09What we are watching for
- Anthropic’s model documentation and release notes, read on every scan.
- Provider catalogs (Bedrock, Vertex, Foundry, OpenRouter) for any new claude-opus identifier.
- A checkable artifact behind the routing claim: an early-access string, a model-list screenshot, a traffic report.
- @notjazii revising the date; the Thursday or Friday window closes September 18.
- Anthropic saying anything at all about what serves Claude Code.
§ 10The tracker
This page updates when facts change, not when posts get louder. As of September 14, 2026: Opus 5.2 has been described by one X account from inside its own Claude Code sessions, repeated by a dozen more, priced by a prediction market, and mentioned by nobody at Anthropic. No date, no spec, no identifier, no card exist in any Anthropic document. The most likely picture is a real next Opus arriving on Anthropic’s schedule, not the rumor’s, and when Anthropic moves the confirmed facts land here the same day next to a graded scorecard of every claim above.
§ 11Update log
This is a living page; when the story moves, the update lands here.
As of September 14, 2026, 16:15 UTC: page opened. The rumored window is September 17 or 18 from the source account, end of September from two others; no official movement to log.
§ 12Sources
- @notjazii on X, read directly: 07:57 UTC, 12:15 UTC, 14:00 UTC, 14:38 UTC, 15:53 UTC, all September 14, 2026. @goodworse, 15:47 UTC and 15:52 UTC. @pradeepXkapoor, September 13. Tim Hua, March 20, 2026. Elon Musk, 11:19 UTC September 14. Times from the posts’ own timestamps.
- Anthropic, Claude Opus 5, July 24, 2026; Claude Fable 5.1 and Mythos 5.1, September 1, 2026; Claude Fable 5 and Mythos 5, June 9, 2026; Claude Sonnet 5, June 30, 2026.
- Anthropic, models overview, Claude Opus 5 model page, platform release notes, pricing, all read September 14, 2026, 16:10 UTC.
- Polymarket, Next Claude Opus released by…?, read September 14, 2026, 16:06 UTC.
- CellCog routing statements reflect our configuration on September 14, 2026.
Q1Has Anthropic announced Claude Opus 5.2?
No. We read Anthropic’s model documentation, the platform release notes, the news index, the status page and the Anthropic catalog on OpenRouter on September 14, 2026. None contains Opus 5.2 or a claude-opus-5-2 identifier. The documentation lists four current models: Claude Fable 5.1, Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5.
Q2What is the tibo test?
A prompt circulating with the rumor: ask Claude Code whether it knows who tibo, the reset guy, is, without searching. If the model answers, the posts say you have Opus 5.2. The prompt tests whether the model knows something recent. A newer checkpoint would pass it; so would a model given the fact by retrieval or a system prompt, and a model that learned it late in training. It cannot tell a version number from a knowledge date. Anthropic’s model page dates Opus 5’s training data to May 2026.
Q3Has Anthropic quietly routed models before?
Something like it, yes. The August rumor about Opus 5.1 rested on early-access identifiers seen in third-party model lists and on reported API traffic for a test string, with some requests falling back to Opus 4.8. Those were artifacts a reader could check. The Opus 5.2 claim has none: no identifier, no traffic report, no screenshot of a model list. It has answers that one user found cleaner and faster.
Q4Did Elon Musk confirm Opus 5.2?
No. His September 14 post grades Grok 4.7 as roughly on par with Opus 5.0, not 5.1, and says nothing about a 5.2. Anthropic has never shipped an Opus 5.1; the 5.1 that exists is Claude Fable 5.1. His post is part of why the name was trending when the routing claim landed, not a source for it.
Q5How often does Anthropic ship a new model?
In 2026 so far: Claude Fable 5 on June 9, Claude Sonnet 5 on June 30, Claude Opus 5 on July 24 and Claude Fable 5.1 on September 1, gaps of 21, 24 and 39 days. An Opus 5.2 on September 18 would land 17 days after Fable 5.1 and 56 days after Opus 5; on September 30, 29 and 68 days. The first would be the shortest gap of Anthropic’s year.
Q6Should I wait for Opus 5.2?
Not on this evidence. Claude Opus 5 is live at $5 per million input tokens and $25 per million output with a 1M-token context, and a 5.2 would arrive as a new identifier when it ships. If you are choosing where your agents run rather than which API to call, the model matters less than the layer around it: memory that survives the conversation, a task board, an inbox, approvals. CellCog’s Creative tier runs on Opus 5 today, Core and Max on Fable 5.1, and a standing AI employee keeps its role across every engine swap. Try it free, no credit card needed; you pay for the work, not the hire, and the cost depends purely on how much work you assign.