Announced but not yet available
Confirmed by vendor
Mistral Large 4 Mistral AI
When: 29 October 2026, the estimate on Mistral's official Hugging Face page
Mistral promised to publish the weights of its largest model by the end of October. Its official Hugging Face page lists 29 October 2026 as the expected date.
- When Mistral launched the model on 6 October 2026, it said the weights would follow by the end of the month. Until then, it is only available as a preview through the API and Mistral Studio.
- Mistral's official Hugging Face repository is marked as an upcoming release and lists 29 October 2026 as the current estimate.
- Mistral has not announced the licence for the weights yet. The Hugging Face date is only an estimate and may still move.
Related model in the catalog: Mistral Large 4
Reported October 28, 2026 · Mistral AI: Mistral Large 4 ↗ · Mistral AI na Hugging Face: Mistral-Large-4.0-1T05-A52B ↗
Confirmed by vendor
Kimi K3.1 Pro Moonshot AI
When: by 4 November 2026 (announced by Moonshot)
Moonshot AI launched Kimi K3.1 Pro in its API on 23 October 2026 and promised to publish the full weights by 4 November. Companies could then run it on their own servers.
- The Kimi documentation says the full model weights will be released by 4 November.
- Moonshot has not said which licence the weights will use.
- According to Let's Data Science, the model has been available through Moonshot's API since 23 October 2026 with a 1 million token context.
Related model in the catalog: Kimi K3.1 Pro
Reported October 23, 2026 · Moonshot AI: Kimi K3.1 Pro Quickstart ↗ · Let's Data Science: Moonshot Releases Kimi K3.1 Pro API, Open Weights Due Nov 4 ↗
Confirmed by vendor
Gemini 4 Carbon Google
When: no date yet, Google says “soon”
Google announced the model on 28 October 2026, but so far only selected cyber defenders have it. Google wants to open it to developers, businesses and consumers soon, without giving a date.
- Google said a subset of cyber defenders gets Carbon this week, with more trusted partners added gradually through the Fairwind Program.
- Before opening it to developers, businesses and consumers, Google says it wants to gather feedback and keep refining the safeguards.
- Google set an introductory price of $2 per million input tokens and $10 per million output tokens. After that, Google says the price rises to $4 and $20.
Related model in the catalog: Gemini 4 Carbon
Reported October 28, 2026 · Google: Introducing Gemini 4 Carbon ↗
Confirmed by vendor
Muse Wonder Meta
When: no date yet, Meta says “soon”, subscribers first
Meta has announced that Muse Wonder, one of its two most capable models, will soon power features in the Muse app as well. Subscribers get it first, and Meta has not given a date.
- On 12 October 2026, Meta said Muse Wonder is available to developers through the Meta Model API and Muse Code. The larger Muse Wonder Full is limited to Muse Code subscribers.
- According to Meta, Muse Wonder will soon power features in the Muse app, rolling out to subscribers first.
- Meta also promised a fuller report on its safety testing, again without a date.
Related model in the catalog: Muse Wonder
Reported October 12, 2026 · Meta: Introducing Muse Wonder and Muse Wonder Full ↗
Confirmed by vendor
Claude Mythos 5.1 Anthropic
When: no date yet, Anthropic promises wider access “as quickly as possible”
According to Anthropic, this is the same model as Claude Fable 5.1, reserved for organizations verified through its programs. So far only a set of US organizations has it, with no date for wider access.
- Anthropic's documentation says Mythos 5.1 was released on 1 September 2026 and is only available to organizations verified through programs such as the Cyber Verification Program.
- According to Anthropic's announcement, it is currently available only to a set of US organizations. Anthropic said it is working with the US government to expand access to more domestic and international partners.
- The documentation lists the same price as Fable 5.1, $10 per million input tokens and $50 per million output tokens.
Related model in the catalog: Claude Fable 5.1
Reported September 1, 2026 · Anthropic: Claude Mythos 5.1 (dokumentácia) ↗ · Anthropic: Claude Fable 5.1 and Claude Mythos 5.1 ↗
Confirmed by vendor
Qwen 4 Alibaba (Qwen)
When: no date yet, Alibaba has only said the model is in training
At its Apsara conference in September 2026, Alibaba confirmed it is training Qwen 4. It has not announced a release date, prices or a licence.
- According to Reuters, Alibaba said on 22 September 2026 that it is currently training Qwen 4. The report says later versions, Qwen 4.5 and Qwen 5, are expected to scale to 5 to 10 trillion parameters.
- According to conference reports, Alibaba showed four names on stage: Qwen 4 Max, Plus, Flash and 27B. They are not in its written press release, and none has a price or a date yet.
- According to the OrcaRouter blog, an unofficial estimate of a release in October or early November 2026 circulated on X. Alibaba has not confirmed it.
Related model in the catalog: Qwen3.8-Max
Reported September 22, 2026 · Reuters cez Investing.com: Alibaba plans AI model with 5 trillion to 10 trillion parameters, unveils new chip ↗ · Pasquale Pillitteri: Alibaba unveils Qwen 4 and a 10 trillion parameter roadmap ↗ · OrcaRouter: Qwen 4, The October Window, and Why the vLLM Kernels Matter ↗
Confirmed by vendor
DeepSeek V4.1 Pro DeepSeek
When: no date yet
When DeepSeek released V4.1 Flash in September 2026, it mentioned a V4.1 Pro model that has yet to launch. It has not published a date, prices or specs.
- In its V4.1 Flash announcement of 10 September 2026, DeepSeek said V4 Pro requests would be routed to V4.1 Flash until V4.1 Pro launches.
- According to the OrcaRouter blog, DeepSeek dropped that rerouting after developers objected, so V4 Pro keeps running. The blog says the company has shared nothing more about V4.1 Pro.
Related model in the catalog: DeepSeek V4 Pro
Reported September 10, 2026 · DeepSeek: V4.1 Flash release note ↗ · OrcaRouter: DeepSeek V4.1 Pro Still Unreleased, What Changed in a Week ↗
Rumors and unofficial reports
Rumor, unconfirmedSource credibility: medium
Claude Fable 5.5 (neoficiálny názov) Anthropic
When: no date, leakers on X expected a release a week later in mid October 2026, which did not happen
According to The Information, Anthropic is preparing a new model that should better understand what users actually want. Leaks on social media mention the name Fable 5.5, but it is unclear whether this is the same model. Anthropic has confirmed nothing.
- On 19 October 2026, The Information reported that the model should reason better and infer a user's real goal from a short request. We only know the report from secondary coverage, the original is paywalled.
- According to the report, a person who tested the model says it can spot the wider problem behind a bug and propose a more complete fix. The report does not name the model.
- On 17 October 2026, TestingCatalog described a screenshot from X showing Fable 5.5 in the Azure model catalog. Posts on X claim partners are already testing it, which is unverified.
- As of 28 October 2026, Anthropic has not announced Fable 5.5.
Related model in the catalog: Claude Fable 5.1
Reported October 19, 2026 · Investing.com: Anthropic, Google and OpenAI prepare new AI models, The Information reports ↗ · TestingCatalog: Anthropic prepares to release Claude Fable 5.5 next week ↗
Rumor, unconfirmedSource credibility: medium
Nový vlajkový model OpenAI (bez názvu) OpenAI
When: one or two new models by the end of 2026 according to The Information, the flagship possibly later
According to The Information, OpenAI plans to release one or two new models before the end of the year. A major successor to GPT-6 Astra reportedly still needs safety work and may come later.
- Citing people familiar with the projects, The Information reports that the company will likely release one or two new models before the end of 2026, The Decoder writes.
- It is unclear whether a major new flagship will be among them. The report says that model could still be a ways off.
- The flagship reportedly uses the same looped layer technique as GPT-6 Astra and needs more alignment work so it behaves as expected.
- The cheaper Terra model The Information wrote about on 19 October shipped as GPT-6.1 Terra on 26 October 2026. It is unclear whether it was one of the one or two models mentioned.
Related model in the catalog: GPT-6 Astra
Reported October 14, 2026 · The Decoder: OpenAI will reportedly ship at least one model this year but its flagship still needs alignment work ↗ · Investing.com: Anthropic, Google and OpenAI prepare new AI models, The Information reports ↗ · OpenAI: Introducing GPT-6.1 Terra ↗
Rumor, unconfirmedSource credibility: medium
DeepSeek V5 (pracovný názov) DeepSeek
When: no date, AIBase expects November 2026 rather than October
According to The Information, DeepSeek is training a model with about 2 trillion parameters. References to V5 reportedly appeared in DeepSeek's code on 27 October. It is unclear whether this is the same model, and the company has announced nothing.
- In September 2026, The Information reported that DeepSeek is training a new model with about 2 trillion parameters. It did not name the model, 36Kr writes.
- According to AIBase, the names DeepSeek-V5, V5-open and V5-next appeared in DeepSeek's Harness codebase on 27 October 2026. It thinks a release is unlikely in October and more likely in November at the earliest.
- DeepSeek has not officially said anything about V5.
Related model in the catalog: DeepSeek V4 Pro
Reported October 28, 2026 · KuCoin News (AIBase): DeepSeek V5 model code appears, suggesting a November launch ↗ · 36Kr: The Information o tréningu modelu s 2 biliónmi parametrov ↗
Rumor, unconfirmedSource credibility: medium
Grok 4.8 xAI
When: no date, Musk mentioned the model in September 2026
In September 2026, Elon Musk wrote on X that Grok 4.8 has 2.5 trillion parameters and will be a noticeable improvement. xAI has not released it or given a date.
- On 13 September 2026, Musk wrote on X that Grok 4.8, a 2.5 trillion parameter model, would finish training that week and move on to reinforcement learning.
- A day later he added that Grok 4.8 will be a noticeable improvement. The parameter count comes only from Musk, xAI has not published it.
- According to the CellCog blog, the name grok-4.8 appeared in late September in the code of an open source tool that works with the Cursor editor. That is a third-party trace, not an announcement.
- As of 28 October 2026, Grok 4.8 is not listed in xAI's API release notes.
Related model in the catalog: Grok 4.7
Reported September 28, 2026 · CellCog: Grok 4.8 Release Date, The 2.5T Model Musk Just Named ↗ · Elon Musk na X: Grok 4.8 dokončuje tréning ↗ · xAI: API release notes ↗
Rumor, unconfirmedSource credibility: medium
GLM-6.0 Z.ai (Zhipu AI)
When: no date, the A2Agent blog estimates early 2027
Zhipu (Z.ai) founder Tang Jie has publicly mentioned a future model, GLM-6.0. He says self-improvement will be one of its directions. The company has not given a release date.
- According to the Chinese portal 10jqka, Tang Jie named self-improvement as one direction for the future GLM-6.0 at a half-year results briefing on 31 August 2026.
- He said the biggest challenge is for the model to tell on its own when to stop and when to fix its own mistakes, which matters more than simply making the model bigger.
- As of 22 October 2026, the A2Agent blog says GLM-6 has no model card, price or weights. It estimates a release in early 2027, which is only a guess.
Related model in the catalog: GLM-5.3
Reported September 1, 2026 · 10jqka: Zhipu o ďalšom základnom modeli GLM-6.0 ↗ · A2Agent: GLM-6, Release Date Predictions and What to Expect vs GLM-5.3 ↗
Rumor, unconfirmedSource credibility: low
Gemini 4 Flash (pracovný názov) Google
When: no date
The name gemini-4-flash-preview reportedly appeared in a leaked screenshot of an API configuration. Google has not announced a cheaper Gemini 4 model.
- On 23 October 2026, NokiaPowerUser wrote about a screenshot showing the identifiers gemini-4-flash-preview and gemini-4-carbon-preview. It did not say who obtained it.
- Google did announce Gemini 4 Carbon from the same screenshot on 28 October. It has said nothing about a Flash model so far.
Related model in the catalog: Gemini 3.8 Flash
Reported October 23, 2026 · NokiaPowerUser: Gemini 4 Flash & Gemini 4 Carbon Leaked, What We Know So Far ↗
Rumor, unconfirmedSource credibility: low
Grok 5 xAI
When: no date, Musk's earlier Q1 2026 target has passed
Musk keeps hinting that Grok 5 will be a big leap, but xAI has never published a date and Musk's earlier estimates did not hold. Musk has also been talking about Grok 4.8 and 4.9.
- In January 2026, xAI officially said Grok 5 was in training. Musk had earlier talked about a release in the first quarter of 2026, and that date has passed.
- On 14 September 2026, Musk wrote on X that Grok 4.8 will be a noticeable improvement, Grok 4.9 is probably Astra or Fable class and Grok 5 may be better than anything.
- On 24 September, Musk wrote that he was cautiously optimistic SpaceX, which xAI is part of according to zBuild, would have a Fable or GPT-6 level model in two to three months. He did not say which model.
Related model in the catalog: Grok 4.7
Reported September 24, 2026 · zBuild: Grok 5 Release Date and Status (October 2026 Update) ↗ · Elon Musk na X: Grok 4.8, 4.9 a Grok 5 ↗