Meta's Watermelon released by October 31, 2026?

- [2026-07-02/05] At an internal Meta Superintelligence Labs town hall, CAIO Alexandr Wang told employees Meta's next large model, codenamed Watermelon, has caught up to OpenAI's GPT-5.5 across several key benchmarks , and that "Watermelon, our next model after Avocado, is currently in training," using "an order of magnitude more compute than Avocado," referring to the Muse Spark codename. No independent verification exists. - [2026-08-25] The Information, citing internal Meta documents, reported Meta Platforms plans to launch its consumer version of the OpenClaw AI agent, dubbed Hatch internally, as soon as the next several weeks and is targeting October for its latest AI model, called Watermelon ; the documents did not clarify whether Watermelon becomes part of the Muse model family or ships as a standalone model . - [2026-09-02] Meta released Muse Spark 1.3 (API-only, not open-weight) and on X, AI at Meta's sign-off said Stay tuned for more updates soon, including bigger models, Muse Spark open weights, and more ; Meta's own research blog closes with the identical roadmap line, with an exciting roadmap lined up, including bigger models, the Muse Spark open weights release, and more — no date or model name given. - [2026-09-02] Press analysis of the tease: the watermelon emoji is the public face of Meta Watermelon — a working codename for Meta's next frontier-scale model, reported since July to be in training at roughly ten times the compute of its predecessor , but Meta itself has not confirmed the mapping. - [2026-09-19] Meta's Muse Connectors developer platform opened, ahead of Meta Connect, per the OrcaRouter retrospective on the keynote. - [2026-09-23/24] Meta Connect 2026 keynote ran in Menlo Park and delivered no Watermelon launch: a detailed retrospective states Meta Watermelon was supposed to be the reason to watch Meta Connect 2026... the model... was not in it. No new frontier model was announced at all ; instead Meta announced Muse Realtime Avatar, Muse Voice, and hardware (Ray-Ban Meta glasses, VR glasses), per the same source. - [2026-09-24] As of this date, no Muse Spark 1.2 repository exists on Hugging Face , 45 days after the August 10 open-weights promise, indicating Meta is behind on even its lesser roadmap commitment, let alone Watermelon. - [2026-09-24] The same report notes a secondhand, undated claim that Wang saying Meta would soon release its most capable model yet, with the shipped Muse Spark line framed as an "appetizer" — exec-stated, not a product announcement, and not independently dated. - [2026-09-24] Per reporting cited in the retrospective, CNBC has since reported that Meta paused a pre-training stage in July and later resumed it, pushing the launch to no earlier than October , and the piece concludes that distribution (open vs. closed) and lineage (Muse-family vs. standalone) remain unknown. - [2026-09-29/28] Rival frontier launches continued in the same period (e.g., OpenAI GPT-6.1 Sol Sept 29, Anthropic Claude Sonnet 5.5 Sept 28), underscoring competitive pressure but with no bearing on Watermelon's own timeline, per model-tracker listings on orcarouter.ai. - [search, no later update found] No search as of October 4, 2026 turned up any confirmation, preview, waitlist, or official Meta announcement of Watermelon's release, name change, or specific launch date; all October-timing claims trace back to the single August 25 The Information report. BOTTOM LINE: As of October 4, 2026, Watermelon remains an unreleased, unconfirmed internal codename with only a reported (not official) "October" target; the one concrete pre-deadline checkpoint (Meta Connect, Sept 23–24) passed with no model reveal and even a smaller related promise (Muse Spark 1.2 open weights) is already overdue, suggesting slippage risk is high for an Oct 31 public release. NOT FOUND: Any official Meta announcement, product page, waitlist, or API listing for "Watermelon"; any confirmed final product name; any specific release date narrower than "October"; any news dated after September 24, 2026 about Watermelon's status.

- [2025-04-05] Meta's prior "next-gen flagship" promise (Llama 4 Behemoth) was announced as "still in training" alongside Scout/Maverick, then has been reported to have its planned release date postponed until fall 2025 or later , and ultimately Behemoth was never publicly released, serving instead as a teacher model before Meta's subsequent launch of Muse Spark in April 2026 marked a strategic shift away from the open-weight frontier approach . https://www.thundercompute.com/blog/llama-4-ai-model-ollama - [2026-05-16] Meta's most recent flagship ("Avocado"/Muse Spark) also slipped internally: the model had been internally targeted for a first-quarter launch, but Meta pushed the rollout back to at least May—and possibly June—after internal tests showed it lagging behind rivals . https://www.aol.com/articles/avocado-ice-meta-afford-pause-153052827.html - [2026-07-02] The "Watermelon" codename surfaced via an internal town hall where Chief AI Officer Alexandr Wang delivered an announcement to employees that Meta's next flagship model, codenamed Watermelon, had matched the performance of OpenAI's GPT-5.5 on closely watched benchmarks ; the claim is internal/unverified and the model was still training. https://www.twinstrata.com/metas-secret-watermelon-ai-model/ - [2026-08-25] The first concrete timing report came from The Information: Meta Platforms will launch its AI agent Hatch in the coming weeks and release a new AI model called Watermelon in October, according to internal documents seen by The Information . https://the-decoder.com/metas-paid-ai-agent-hatch-launches-soon-with-a-new-model-called-watermelon-due-in-october/ - [2026-09-02] Meta's own official roadmap language (the resolution source cited in the market) was vague, not a date: We have an exciting roadmap lined up, including bigger models, the Muse Spark open weights release, and more . https://research.meta.ai/blog/introducing-muse-spark-1-3 - [2026-09-06] Independent analysis separated confirmed facts from rumor, noting Meta Watermelon is a working codename, not a product name Meta has confirmed, and what is known comes from three layers of reporting, none of them an official model card . https://www.orcarouter.ai/blog/meta-watermelon-leak - [2026-09-23] A key base-rate data point: Meta's own developer conference (Meta Connect) passed without a Watermelon reveal — the piece reports the keynote "was not in it. No new frontier model was announced at all," and relays that CNBC reported a pre-training pause that "pushed the launch to no earlier than October" (per the fetched orcarouter.ai article, unverified/press-sourced). - [2026-09-24] The companion promise bundled with Watermelon in the same Zuckerberg post — open-weighting Muse Spark 1.2 — also remains unmet: the same source states that "forty-five days after the August 10 promise... no Muse Spark 1.2 repository exists on Hugging Face," illustrating a pattern of slipped Meta AI roadmap commitments (per fetched orcarouter.ai article). - [2025-05 to 2026-07] Historical reference class: Meta's last comparably-hyped "frontier" model (Behemoth) went from announced/teased to permanently unreleased over more than a year, a track record several trackers describe as "shelved" with mid-training MoE-routing and chunked-attention issues at 2T-scale meant Meta lost confidence the gains justified shipping . https://codersera.com/blog/llama-4-complete-guide-2026/ - [2026-08-25] Multiple outlets (Yahoo Finance, Stocktwits, tech-insider.org, kie.ai) converged on the same single primary leak (The Information) rather than independent confirmation, meaning the "October" date is a single-sourced report amplified, not a confirmed Meta commitment: these tools have been central to our ongoing coverage of Meta's AI trajectory... currently targeting an October 2026 release, internal claims suggest Watermelon has achieved GPT-5.5 parity on internal benchmarks . https://finance.yahoo.com/technology/ai/articles/meta-hatch-agent-platform-watermelon-113350203.html BOTTOM LINE: Meta's own reference class (Llama 4 Behemoth never shipped; Muse Spark/"Avocado" slipped a full quarter) shows that Meta's internally-teased "next frontier model" announcements have a poor and slow track record of turning into fast, on-schedule public releases, and the single-sourced "October 2026" Watermelon target has already seen one flagship venue (Meta Connect, Sept 23-24) pass without a launch plus reports of a mid-cycle training pause — all of which argue for meaningful base-rate skepticism about a full public release inside the Oct 31 window. NOT FOUND: No official Meta statement committing to a specific Watermelon release date; no Polymarket/Kalshi-specific odds history beyond the single tweet referencing the same Information report; no confirmation of whether "Watermelon" will be open-weight, API-only, or consumer-app-only, which affects whether a release would meet the contract's "generally available" bar.

YES means: Meta publicly releases "Watermelon" (or the model credibly identified as that internal codename) with general public access — open beta, waitlist, or GA — on or before October 31, 2026 (ET). - [2026-08-25] The primary basis for a YES case is The Information's internal-documents report that Meta Platforms plans to launch its consumer version of the OpenClaw AI agent, dubbed Hatch internally, as soon as the next several weeks and is targeting October for its latest AI model, called Watermelon , a month-specific internal target that still has four weeks of runway left inside the window. https://www.theinformation.com/articles/meta-plans-launch-hatch-ai-agent-platform-coming-weeks - [2026-09-03] A new, more recent on-the-record data point not previously logged: the day after Zuckerberg's public tease, CAIO Alexandr Wang told staff Watermelon's development is on track, stating Watermelon's development is on track and that "we believe Watermelon will be extremely competitive." This is a post-tease executive reaffirmation, not a walk-back or delay signal. https://www.orcarouter.ai/blog/meta-watermelon-leak - [2026-09-06] Reporting ties Watermelon's reported pause directly to a resumed schedule rather than an indefinite slip: per the same analysis, "Reported target — October 2026, per The Information's August 25 report citing internal documents; CNBC's pre-training-pause reporting points the same way." The pause narrative itself still lands the model in October, not past it. https://www.orcarouter.ai/blog/meta-watermelon-leak - [2026-09-06] Infrastructure signal: the same source reports the training cluster is Meta's own purpose-built, large-scale compute, noting Compute — "an order of magnitude" more than the previous generation, per Wang's July 2 all-hands as reported; not independently verified. The training cluster is reported to be Prometheus, Meta's roughly 1-gigawatt New Albany, Ohio site, at around 500,000 GPUs. A dedicated gigawatt-scale cluster indicates committed, well-resourced training rather than a stalled side project. https://www.orcarouter.ai/blog/meta-watermelon-leak - [2026-09-02] Meta's own CEO-level public roadmap tease (not just leaks) explicitly flagged an imminent "bigger model," per the company's research blog: On August 25, 2026, BlockBeats (citing The Information and Meta internal documents) reported that Meta would ship HATCH — a consumer version of its OpenClaw AI agent — within weeks, and that its latest AI model, Watermelon, is slated for an October release. Meta choosing to publicly tease the model with an emoji on its own flagship launch post is evidence of an active, near-term release intent rather than a project being quietly shelved. https://www.eggstriker.com/en/blog/meta-watermelon-model-2026 - [2026-08-30] Multiple independent outlets converged on the same October target drawn from internal planning documents rather than speculation, with one analysis noting Both Investing.com and Benzinga report that Meta is targeting an October 2026 release for the model, with LLMs.blog describing it as Meta's next flagship foundation model that would eventually become the primary inference backend behind both Hatch and Meta AI. Convergent sourcing on a specific month (rather than vaguer "coming months" language) raises the credibility of an October ship intent. https://shattered.io/meta-hatch-ai-agent-watermelon-model-2026/ - [2026-08-25] The low bar set by the resolution criteria favors YES: the market only requires "open beta or open rolling waitlist signups," not GA pricing or full rollout; reporting indicates Meta is willing to ship a less-than-fully-finished product first and iterate, as it already does per Rather than waiting for Watermelon to be ready before shipping an agent product, Meta appears willing to launch Hatch first and swap in the more powerful model once it clears internal testing. This "ship fast, iterate" posture from Meta Superintelligence Labs increases the odds of at least a waitlist-level Watermelon release landing inside October. https://tech-insider.org/meta-hatch-ai-agent-watermelon-model-2026/ - [2026-09-06] Meta's recent cadence under CAIO Wang has been unusually fast: the same tracker notes the run-up included On August 5th, Meta released Muse Spark 1.2 and the Muse Code terminal agent. Five days later, Meta published Muse Glimmer, a 30-billion-parameter open-weight model designed for local agent workflows. This four-model-in-four-months cadence under Meta Superintelligence Labs is a concrete, checkable base rate of rapid, frequent shipping that supports Meta hitting another release inside the Oct 31 window, even if the "Watermelon" label or full lineage remains unsettled. https://runtimewire.com/article/meta-hatch-ai-agent-watermelon-model-launch BOTTOM LINE: The strongest case for YES rests on a specific, internally-sourced "October" target reaffirmed by Wang on Sept 3, a dedicated ~500,000-GPU training cluster signaling serious resourcing, Meta's own public roadmap tease, and a demonstrated pattern of rapid, lower-bar (open-weight/waitlist) shipping cadence from Meta Superintelligence Labs — all of which leave real runway for an open-beta-level Watermelon release before October 31. NOT FOUND: No official Meta announcement, product page, or waitlist for Watermelon as of this search; no confirmation that Hatch itself has fully launched to the general public (reports only as of late Aug/Sept describe it as imminent); no specific calendar date for Watermelon narrower than "October"; no post-September 24 news confirming continued on-track status beyond the Sept 3 Wang remarks already cited.

YES means: Meta publicly releases "Watermelon" (or the model confirmed to be the internally codenamed Watermelon model, under whatever final name) with general public access — open beta, open waitlist, or full GA — on or before October 31, 2026 (ET). The NO case argues this does not happen by that date. - [2026-08-25] The only concrete timing claim for Watermelon traces to a single primary source: "Meta Platforms will launch its AI agent Hatch in the coming weeks and release a new AI model called Watermelon in October," according to internal documents seen by The Information , and even that report could not establish distribution or lineage: "Whether Watermelon becomes part of the Muse model family or ships as a standalone model is still unclear." https://the-decoder.com/metas-paid-ai-agent-hatch-launches-soon-with-a-new-model-called-watermelon-due-in-october/ - [2026-09-06] Independent analysis stresses that even the "October" date itself is unverified, not a Meta commitment: "Meta Watermelon is a working codename, not a product name Meta has confirmed. What is known comes from three layers of reporting, none of them an official model card." https://www.orcarouter.ai/blog/meta-watermelon-leak - [2026-09-06 (updated to reflect Sept 23-24)] The retrospective version of this same tracker confirms the single scheduled pre-deadline checkpoint came and went with nothing: "Meta Watermelon was supposed to be the reason to watch Meta Connect 2026... the model... was not in it. No new frontier model was announced at all." https://www.orcarouter.ai/blog/meta-watermelon-leak - [dated in same retrospective] A named mechanism for slippage beyond just "reported": the piece states CNBC reported a mid-cycle training interruption, concluding " Third, CNBC has since reported that Meta paused a pre-training stage in July and later resumed it, pushing the launch to no earlier than October. " This means even the optimistic case places the earliest plausible date at the very start of the 8-week window, leaving zero slack for safety testing, infra issues, or scope changes before Oct 31. - [2026-09-24, per fetched piece] A directly checkable companion commitment from the same Sept 2 post — Muse Spark 1.2 open weights — remains unmet 45+ days after being promised "in the coming weeks," and the piece states plainly: "no Muse Spark 1.2 repository exists on Hugging Face" ; this is a same-order-of-magnitude roadmap promise from the identical announcement that has already blown through its own "soon" language, evidencing Meta's current pattern of execution slippage on this specific roadmap line. - [2026-09-24, per fetched piece] The only non-committal, undated executive remark about imminent capability is explicitly flagged as unverifiable: "Fourth, on September 23, as the keynote ran, coverage relayed Wang saying Meta would soon release its most capable model yet, with the shipped Muse Spark line framed as an 'appetizer' — exec-stated and reported secondhand, though the outlet did not date the remarks, and in any case not a product announcement." - [2026-09-24, per fetched piece] Even the lineage/branding question needed for resolution remains unresolved by Meta itself: "Lineage — unclear: Muse-family successor, Llama-family release, or standalone; internal documents reportedly do not say." and "Distribution — unknown, including whether it will ship open-weight. Nothing at Connect changed that." This is directly relevant to the resolution criteria's requirement that the model be "identified... as the model internally codenamed 'Watermelon'" and "made available to the general public" — an undecided open-vs-closed distribution path adds a further gating step (a public-access decision) that must still occur before Oct 31. - [45-day pattern check, 2026-09-24] The same source independently verified via Hugging Face that Meta's org page shows "A direct check of Meta's model organisation on the Hub today returns four repositories, all of them Muse Glimmer variants: there is no Muse Spark checkpoint of any version, no license, no hardware spec and no date." This is a concrete, checkable data point (not just reporting) showing Meta missing its own smaller, easier-to-fulfill roadmap promise from the same announcement that teased Watermelon. - [cross-checked 2026-10-04] No search turned up any post-September 24 development — no waitlist, no product page, no leaked release notes, no confirmed name change — beyond reiterations of the same August 25 The Information report being recycled by secondary outlets (Yahoo Finance, Stocktwits, PYMNTS, tech-insider.org, eggstriker.com, shattered.io) through late September, indicating no forward momentum signal has emerged in the final month of the window. - Base-rate mechanism: Meta's own historical pattern on "next frontier model" promises (Llama 4 Behemoth never shipped; Muse Spark/Avocado itself slipped a full quarter past its internally targeted window) shows that internally-sourced "X month" targets for Meta's largest training runs have repeatedly proven unreliable, and Watermelon — reported to be using roughly 10x the compute of its predecessor — is Meta's largest and most technically ambitious training run yet, raising (not lowering) slippage risk relative to prior cycles. - Resolution-criteria technicality: the market requires the release be "publicly accessible," explicitly excluding "a closed beta or any form of private access," and with distribution still undecided per the above, even if Meta announces internal readiness by Oct 31, a closed/invite-only soft launch (as it did with the original Muse Spark debut) would not satisfy the resolution criteria, adding a second independent hurdle beyond mere technical completion. BOTTOM LINE: Every concrete timing claim for Watermelon reduces to one single-sourced, unconfirmed "October" target from an August 25 report; the lone scheduled pre-deadline showcase (Meta Connect, Sept 23-24) delivered no model and no distribution clarity, a smaller companion promise from the same announcement is already over six weeks overdue, and Meta's own track record (Behemoth, Avocado) shows frontier-model timelines slipping routinely — all of which argue strongly that a public, access-qualifying Watermelon release within the remaining ~4 weeks to Oct 31 is unlikely. NOT FOUND: Any Meta-official statement, product page, waitlist signup, or API listing for Watermelon; any news dated after September 24, 2026 updating Watermelon's status; any confirmation of open-vs-closed distribution; any specific release date narrower than "October."