Meta's Watermelon released by October 31, 2026?

- [2026-10-12] Nothing I found shows Watermelon released. The resolution source named on Polymarket's page, the Muse Spark 1.3 post (published 2026-09-02), still lists Muse Spark 1.3 as the newest model. Its only forward-looking text is "bigger models, the Muse Spark open weights release, and more. Stay tuned." It gives no model name and no date. https://research.meta.ai/blog/introducing-muse-spark-1-3 - [2026-10-12] Polymarket's event page says no official announcement, weights or external benchmarks have emerged, and that the model was absent from the September Meta Connect keynote. https://polymarket.com/event/metas-watermelon-model-released-byptptpt - [2026-09-02] Meta released Muse Spark 1.3 on Muse Code and the Meta Model API. Muse Spark 1.3 with max reasoning came later; the Sept 25 update to Fello AI's review says it went generally available on Sept 4. AI at Meta's X post said "bigger models" and Muse Spark open weights were coming. https://x.com/AIatMeta/status/2095234385129963666 - [2026-09-02] Zuckerberg's launch post reportedly ended with "Next up 🍉 and Muse Spark open weights releases coming soon." The watermelon-emoji link to Watermelon is an inference by commentators, not a Meta statement. https://felloai.com/muse-spark-1-3/ - [2026-08-25] The Information, citing internal documents, reported that Meta is targeting October for Watermelon. The documents did not say whether it would sit inside the Muse family or ship standalone. The date comes from leaks, not from Meta. https://the-decoder.com/metas-paid-ai-agent-hatch-launches-soon-with-a-new-model-called-watermelon-due-in-october/ - [2026-09-03] Wang reportedly said Watermelon's development is on track and that it will be "extremely competitive." He gave no date. This comes from a secondary blog. https://www.orcarouter.ai/blog/meta-watermelon-leak - [2026-09-23] Meta Connect ran Sept 23–24 and announced no new frontier model. Per OrcaRouter, announcements were Muse Realtime Avatar (due "in the next months"), Muse Voice and hardware. The same page says Wang described the shipped Muse Spark line as an "appetizer" and said a more capable model would come "soon." That remark is secondhand and undated. https://www.orcarouter.ai/blog/meta-watermelon-leak - [2026-09-06] The same blog cites CNBC reporting that Meta paused a pre-training stage in July, then resumed it, pushing launch to no earlier than October. I could not find the CNBC article itself, so treat it as unverified. https://www.orcarouter.ai/blog/meta-watermelon-leak - [2026-07-09] Wang said Meta was training Watermelon and gave no release date (CNBC). In a July 2 town hall (Business Insider, via secondary sources), he said it was still in training, used about 10x the compute of Avocado/Muse Spark, and had reached GPT-5.5 parity on internal benchmarks. https://www.cnbc.com/2026/07/09/meta-jumps-into-ai-coding-market-to-chase-anthropic-and-openai.html - [2026-10-11] Lines.com, citing the Polymarket market, reported a sibling date-strike structure. The "Oct 7" strike was about 12% and the "Oct 21" strike about 11% on a snapshot from Oct 3–7. The page's headline figure is 23% as of Oct 11, but it is ambiguous which strike that refers to. Volume is thin (about $28K). The Oct 7 strike has now lapsed with no release. https://www.lines.com/prediction-markets/tech/ai/metas-watermelon-released-onptptpt-20260928142913437 - [2026-08-25] The Hatch consumer agent was reported as due "in coming weeks," with Watermelon to follow in October. I found no evidence that Hatch has launched. https://thenextweb.com/news/meta-hatch-ai-agent-watermelon-199-subscription - [2026-10-12] Meta's cadence this year: Muse Spark on Apr 8, 1.1 on Jul 9, 1.2 on Aug 5, 1.3 on Sep 2, and Muse Glimmer 30B (open weights) on Aug 10. Past slips: the Avocado/Muse Spark launch was delayed per NYT/NY Post, and the promised Muse Spark open weights have not appeared. https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model BOTTOM LINE: As of Oct 12 Watermelon is unreleased and unannounced by Meta, and the only dated target (October) comes from leaked internal documents. About 19 days remain, and Meta has given no on-record date. A qualifying public launch would need to be announced within that window. NOT FOUND: Any Meta statement or teaser dated after Sept 24 about Watermelon timing. The CNBC article on the pre-training pause. Any news from the last 7 days beyond market snapshots. Any scheduled Meta event between Oct 12 and Oct 31 (Q3 earnings is likely late October, but I did not confirm the date). Any launch of Hatch.

- [2026-07-02] Business Insider reported that Alexandr Wang told staff at an internal town hall that Watermelon, still in training, matched GPT-5.5 on internal benchmarks. It uses about 10x the compute of Avocado (Muse Spark). No release timeline was given. https://yowox.com/posts/meta-watermelon-gpt-5-5-benchmark-claim/ - [2026-08-25] The Information, citing internal documents, reported Meta is targeting an October release for Watermelon. Hatch was to come first. The documents did not say whether Watermelon would sit inside the Muse family or ship standalone. https://the-decoder.com/metas-paid-ai-agent-hatch-launches-soon-with-a-new-model-called-watermelon-due-in-october/ (the page itself failed to fetch; the same facts appear in search snippets at https://runtimewire.com/article/meta-hatch-ai-agent-watermelon-model-launch) - [2026-08-30] Aggregators note the October date does not come from Meta. Meta has not confirmed the codename, the date or any specs. https://shattered.io/meta-hatch-ai-agent-watermelon-model-2026/ - [2026-09-02] Meta's Muse Spark 1.3 post, which I fetched in full, says only "bigger models, the Muse Spark open weights release, and more. Stay tuned." It gives no name, date or Watermelon reference. https://research.meta.ai/blog/introducing-muse-spark-1-3 - [2026-09-02] Meta's cadence: Muse Spark on 2026-04-08, 1.1 on 07-09, 1.2 on 08-05, 1.3 on 09-02. That is roughly one release every 4 weeks since July, which is fast. The first gap was 3 months, and each was an incremental version. The next 4-week slot lands around 09-30 to 10-07. As of today I found no sign of a new release. https://ai-tldr.dev/models/muse-spark-1-3/ - [2026-09-02] Even 1.3 was staged. Max reasoning was in limited partner preview, with broad rollout "shortly" after safety testing. A flagship could likewise ship first in limited or partner access, which would not qualify under the "no closed beta" rule. https://venturebeat.com/technology/meta-says-muse-spark-1-3-has-frontier-performance-but-its-best-results-come-from-a-model-developers-cant-broadly-use-yet - [~2026-09-02] Reports say Zuckerberg teased a model with a watermelon emoji. That mapping to Watermelon is unverified, and the open-weight Muse Spark promise from 08-10 had not shipped. https://www.orcarouter.ai/blog/meta-watermelon-leak - [Historical] Meta has a record of slipping flagship models. Llama 4 Behemoth was delayed in 2025 over capability concerns. https://en.wikipedia.org/wiki/Meta_Superintelligence_Labs - [2026-10-10] A tech-insider.org piece, with no named source, predicts the October release "will probably slip by several weeks." This is opinion, not reporting. https://tech-insider.org/meta-hatch-ai-agent-watermelon-model-2026/ - [Reference class] Leaked "targeted for month X" dates for frontier models often slip, and a model that arrives on time may still be gated. Watermelon would also need to be identifiable as the codename model, since Meta may ship it as Muse Spark 2 or similar. I did not find a systematic dataset for this. - [2026-10-12] My searches found no report that Watermelon has launched, been delayed, or been given a date. I found no news after about 09-02 specific to Watermelon. BOTTOM LINE: A single leaked "October" target from The Information, with no Meta confirmation and a 3-week window left, sits against Meta's fast ~4-week release cadence and its history of flagship slips. Public availability and the codename-identification requirement add further ways to resolve No. NOT FOUND: Any Meta statement naming Watermelon or giving a date; news from after 09-02 on Watermelon timing; a base-rate dataset for leaked-date accuracy; Polymarket or other venue odds (excluded by instruction); a reliable report on whether the model will launch publicly or in limited access.

YES means: Meta makes Watermelon, or a model credibly identified as the one codenamed "Watermelon" (under any release name), publicly accessible to the general public between now and the 2026-10-31 deadline. Open beta or an open rolling waitlist counts. Closed beta or partner-only access does not. - [2026-10-12] Earlier reports flagged the Q3 earnings date as unconfirmed. Data providers now put Meta's Q3 2026 results on Wed Oct 28, after the close, with the call at 4:30pm ET. MarketBeat lists it, and TipRanks marks it "Confirmed". Meta has used earnings and its own news releases to tie product launches to its AI narrative. A Watermelon launch, or at least a dated announcement, in the Oct 20–28 window would suit that. This is an inference, not a Meta statement. https://www.marketbeat.com/earnings/reports/2026-10-28-facebook-inc-stock/ - [2026-10-12] The Information's original Aug 24–25 report, which I fetched, states the target in plain terms: Meta is "targeting October" for Watermelon. Several outlets syndicated it (Investing.com, PYMNTS, Techstrong, Stocktwits). That is one source repeated, not independent confirmation. The report also made Hatch the first launch, so an October slot is the stated plan rather than a rumor of a slip. https://www.theinformation.com/articles/meta-plans-launch-hatch-ai-agent-platform-coming-weeks - [2026-09-06, as relayed] The OrcaRouter page I fetched carries three further claims. First, the CNBC-sourced pre-training pause "points the same way" as October. Second, Wang said around Sept 23 that Meta would "soon" release its most capable model yet, with the shipped Muse Spark line as an "appetizer". Third, Zuckerberg's sign-off line relayed by @scaling01 was "Next up 🍉 …". The page itself flags the Wang remark as secondhand and undated, and the emoji-to-Watermelon link as unverified. These are the strongest on-record signals that the model is next. They fit an October launch. https://www.orcarouter.ai/blog/meta-watermelon-leak - [2026-09-24] Meta Connect passed with no frontier model. The same page says the model "most likely window for a public reveal has closed" and that the October target is "the next dated thing to watch". Connect only previewed hardware and Muse product features, so a standalone model drop after the event remains open. Meta can still ship a model without a keynote, as it did for Muse Spark 1.1, 1.2 and 1.3. https://www.orcarouter.ai/blog/meta-watermelon-leak - [2026-10-02] Market context raises Meta's incentive to ship. Admirals reports Meta stock rose about 27% in September on Muse momentum, with a 24 Sept high of $779.82 and a close of $725.93 on Oct 1. It cites Muse topping the US App Store. Q3 results are due Oct 28. A new flagship before or at earnings would extend that momentum. Management also faces Q3 capex questions, with 2026 guidance of $130–145B. This is context, not evidence of timing. https://admiralmarkets.com/education/articles/shares/meta-stock-analysis-october-2026 - [2026-09-22] The Shattered.io explainer (updated Sept 22) still repeats the "Watermelon in October" timeline as the most concrete detail in the story. I found no report of a delay, cancellation or new date after Sept 22. Absence of a delay report is weak evidence but is consistent with the plan holding. https://shattered.io/meta-hatch-ai-agent-watermelon-model-2026/ - [2026-10-12] Most likely path to YES, with its prerequisites: - Meta ships Watermelon in the last 10 days of October (roughly Oct 20–30), probably as the next Muse model (e.g. a larger Muse Spark or a new tier). - It lands the way Muse Spark 1.3 did on Sept 2, as a blog post plus availability in Meta AI, Muse Code or the Meta Model API. - Meta, or a consensus of credible reporting, links it to Watermelon. The Information and CNBC have the codename in play, and Wang has talked up "Watermelon" publicly, so identification is plausible. - Access is open at launch: Meta AI app and web, or a self-serve API. The prerequisite most at risk is public access. A staged rollout would fail the test, as 1.3's max-reasoning mode was staged. Meta AI app availability (free consumer access) is Meta's usual route for a headline model. Muse Spark was in the Meta AI app on day one, with only the API in private preview. This is from Meta's April launch post. https://about.fb.com/news/2026/04/introducing-muse-spark-meta-superintelligence-labs/ - [2026-07-09] Cadence supports a fast turnaround. Meta shipped four Muse Spark checkpoints in 5 months (Apr 8, Jul 9, Aug 5, Sep 2). It also shipped Muse Image and Muse Glimmer. The team has repeatedly gone from internal readiness to public availability within days. If Watermelon is in post-training now, public launch inside 19 days is operationally normal for this lab. https://runtimewire.com/article/meta-hatch-ai-agent-watermelon-model-launch BOTTOM LINE: The YES case is a single leaked October target (The Information, Aug 24–25, repeated widely), backed by Wang's "soon" remarks, a fast release cadence, and a plausible Oct 20–28 slot ahead of the Oct 28 earnings call. It needs Meta to hit the target within the last 10 days of the month, ship it openly rather than staged, and have it identifiable as the Watermelon model. NOT FOUND: Any Meta statement after Sept 24 giving a Watermelon date. Any report dated after Sept 22 of a delay or of a launch. The CNBC pre-training-pause article itself. Official Meta confirmation of the Oct 28 earnings date. Any launch of Hatch. Any report of the planned access model (public vs. limited) for Watermelon.

YES means: Meta releases Watermelon, or a model credibly identified as Watermelon under any product name, and makes it publicly accessible by 2026-10-31 ET. Closed beta or partner-only access does not count. - [2026-10-12] ABAB News reports Zuckerberg said in a video interview that Watermelon is a much stronger pre-trained model that "feels good" and will launch "soon". The same item says no model card, price, weights or exact date exists, and that Watermelon was absent from Connect. "Soon" is the only language, with no date. The item has no named outlet or interview, so I could not verify the interview itself. https://www.ababnews.com/news/04b5d575-9dfb-4796-86d6-bc04a64e6e91 - [2026-10-28] Q3 earnings are expected Oct 28 after the close. That leaves only Oct 29–31 after the call. Meta could hold a launch for the call or ship just before it. If Zuckerberg only says "coming weeks" on the call, the odds of a launch before Oct 31 drop sharply. MarketBeat and TipRanks list Oct 28, but the date is unconfirmed on Meta's investor site. https://www.tipranks.com/stocks/meta/earnings - [2026-10-03] A third-party tracker of what Meta has shipped (imisofts, read Oct 3) lists the Muse personal agent as launched Sept 8, with Muse Spark 1.1 and the API on Jul 9. It lists no Watermelon or other flagship-class model. The release list also contains no Hatch launch under that name. The agent product may be Hatch under a new name, which I could not verify. The Information's ordering had Hatch first and Watermelon in October. A model that ships behind the product it is meant to power would sit later in the queue. https://imisofts.com/blog/meta-superintelligence-labs/ - [2026-08-25] The Information's October target was "targeting", from internal documents. It was never a commitment. Investing.com and the other re-reports carry the same single source, so "two independent sources" claims are really one leak. https://www.investing.com/news/stock-market-news/meta-to-launch-hatch-ai-agent-platform-in-coming-weeks-the-information-93CH-4874295 - [2026-08-30] The Hatch precedent: The Information said Hatch would launch "as soon as" late Aug or early Sept. The Muse agent came Sept 8, close to that window, so the leaked dates have not been wildly wrong. But Watermelon's training used about 10x the compute, and the Jul 2 town hall said it was still in training. Safety testing for a larger flagship is the usual gating step. Shattered.io says flagship models slip more often than product launches do. https://shattered.io/meta-hatch-ai-agent-watermelon-model-2026/ - [2026-10-12] The Polymarket page lists the Oct 15 strike at 89% and Nov 30 at 97%. I could not tell whether these are "No" prices or cumulative "by date" prices. Lines.com's Oct 3–7 snapshot gives "No release by Oct 31" at about 22.5%, with the specific dates summing to 77.5%. This is a stale, thin-volume market. The two figures conflict in direction, so they are weak evidence. https://polymarket.com/event/metas-watermelon-model-released-byptptpt - [2026-10-12] Resolution mechanics favour No on ambiguity. A launch must be public (not a limited partner preview), and Watermelon must be identified as the codename model. Meta's Sept 2 Muse Spark 1.3 max-reasoning tier was a partner preview first. If Meta ships "Muse Spark 2" without naming a codename, the match depends on a consensus of credible reporting. Lines.com notes that soft betas and partial API access would not qualify. https://www.lines.com/prediction-markets/tech/ai/metas-watermelon-released-onptptpt-20260928142913437 - [2026-10-12] Even the CEO language is soft. Wang's earlier remark that a coding model near Claude Opus level would come "soon" is also undated. Meta's open-weights promise of Aug 10 is still unmet as of Oct 12, which is evidence that "soon" at Meta has run past eight weeks. https://www.ababnews.com/news/04b5d575-9dfb-4796-86d6-bc04a64e6e91 - [2026-10-10] Tech-insider.org expects the October release "will probably slip by several weeks". It names no source, so this is opinion, not reporting. Its table still lists Watermelon as "reportedly targeted" for October with no update. https://tech-insider.org/meta-hatch-ai-agent-watermelon-model-2026/ - [2025–2026] The prior flagship, Avocado, was expected by late 2025 and slipped to Apr 8, 2026. Cryptobriefing cites this as the key timeline risk for Watermelon. https://cryptobriefing.com/meta-watermelon-ai-model-compute/ - [2026-09-23] The Information's October timing was already six weeks old at Connect. Meta skipped its biggest owned stage for AI announcements. A flagship usually gets a keynote or a dedicated blog post and a pre-announcement. I found no Meta teaser naming a date up to Oct 12. https://polymarket.com/event/metas-watermelon-model-released-byptptpt BOTTOM LINE: With 19 days left, no on-record Meta date, no Connect reveal, only "soon" language and a flagship that was still training in July, the status quo (unreleased) is the default. The earnings call on Oct 28 leaves a short window for a launch, and the public-access and codename-match requirements add further ways to resolve No. NOT FOUND: Any Meta statement after Sept 24 naming a Watermelon date. A named source for the Zuckerberg "soon" interview. Meta's investor-site confirmation of the Oct 28 earnings date. Any report of a Watermelon delay or a launch. Any Watermelon release scheduled for Oct 12–31.