Rumor watch
Last updated Jul 6, 2026
Publicly announced models that are not broadly available yet, limited-access models like Mythos, and more speculative release reports. Rumors stay labeled until a primary source or public release confirms them.
Official signals
Models with a vendor announcement, public preview, or clear launch signal, but not yet a normal broadly available release.
DeepMind - May 19, 2026
Announced at Google I/O 2026; emphasizes deep multimodal reasoning over a 2M-token context. Recent reporting says the broad launch slipped from June toward July while testers continue using it in Google Antigravity and LMArena.
Meta - Apr 5, 2025
Meta's announced but unreleased Llama 4 teacher model: a multimodal MoE with 288B active parameters and nearly 2T total parameters. Meta says it was still training when Scout and Maverick shipped and that those released models were distilled from Behemoth.
Limited access
Models reported as real but gated, suspended, withdrawn, or limited to vetted partners rather than the broader public.
OpenAI - May 7, 2026
OpenAI's limited-preview cybersecurity model for vetted defenders in Trusted Access for Cyber. It is tuned for more permissive authorized security workflows such as vulnerability triage, patch validation, malware analysis, red teaming, and controlled exploit validation, but is not generally available.
Anthropic - Apr 7, 2026
The restricted sibling of Claude Fable 5, sharing the same underlying Mythos-class model with fewer safeguards for vetted defensive-security and research use. Anthropic disclosed Mythos Preview on April 7, 2026, upgraded approved Project Glasswing users to Mythos 5 on June 9, suspended access under a U.S. export-control directive on June 12, and restored limited access for approved U.S. organizations on July 1 while it continues expanding the trusted-access program.
Speculative file
Reports, executive comments, codenames, and launch-window claims that have not yet become public release records.
Mistral - Jul 6, 2026
Rumored / early-access: Mistral AI has confirmed a new open-weight Mixture-of-Experts family — described by CEO Arthur Mensch as "fat but sparse" — aimed at closing the gap with frontier open-weight releases, with early access beginning in July 2026. Mensch confirmed the intent but disclosed almost nothing concrete: no parameter count, no benchmarks, no license terms, and no release date. Tracked as rumored until weights or an official product page land.
Z.ai - Jun 19, 2026
A speculative Z.ai frontier model tracked after Z.ai founder Jie Tang responded to Elon Musk's prediction of a Chinese Fable 5-class model by saying it would not take that long. Name, architecture, weights, and launch timing remain unconfirmed.
DeepSeek - Aug 14, 2025
Rumored successor to DeepSeek R1. Reports say development and launch timing were affected by hardware constraints around Huawei Ascend training and Nvidia availability; final specs, license, and release date remain unconfirmed.
xAI - Aug 8, 2025
Rumored next major Grok model. Elon Musk said after GPT-5's launch that Grok 5 would arrive before the end of 2025, but no broad public Grok 5 release is logged in this catalog yet.
Filter the watchlist
8 models
Rumored / early-access: Mistral AI has confirmed a new open-weight Mixture-of-Experts family — described by CEO Arthur Mensch as "fat but sparse" — aimed at closing the gap with frontier open-weight releases, with early access beginning in July 2026. Mensch confirmed the intent but disclosed almost nothing concrete: no parameter count, no benchmarks, no license terms, and no release date. Tracked as rumored until weights or an official product page land.
A speculative Z.ai frontier model tracked after Z.ai founder Jie Tang responded to Elon Musk's prediction of a Chinese Fable 5-class model by saying it would not take that long. Name, architecture, weights, and launch timing remain unconfirmed.
Announced at Google I/O 2026; emphasizes deep multimodal reasoning over a 2M-token context. Recent reporting says the broad launch slipped from June toward July while testers continue using it in Google Antigravity and LMArena.
OpenAI's limited-preview cybersecurity model for vetted defenders in Trusted Access for Cyber. It is tuned for more permissive authorized security workflows such as vulnerability triage, patch validation, malware analysis, red teaming, and controlled exploit validation, but is not generally available.
The restricted sibling of Claude Fable 5, sharing the same underlying Mythos-class model with fewer safeguards for vetted defensive-security and research use. Anthropic disclosed Mythos Preview on April 7, 2026, upgraded approved Project Glasswing users to Mythos 5 on June 9, suspended access under a U.S. export-control directive on June 12, and restored limited access for approved U.S. organizations on July 1 while it continues expanding the trusted-access program.
Rumored successor to DeepSeek R1. Reports say development and launch timing were affected by hardware constraints around Huawei Ascend training and Nvidia availability; final specs, license, and release date remain unconfirmed.
Rumored next major Grok model. Elon Musk said after GPT-5's launch that Grok 5 would arrive before the end of 2025, but no broad public Grok 5 release is logged in this catalog yet.
Meta's announced but unreleased Llama 4 teacher model: a multimodal MoE with 288B active parameters and nearly 2T total parameters. Meta says it was still training when Scout and Maverick shipped and that those released models were distilled from Behemoth.