Upcoming AI models

At a glance

Status: official = announced or named by the lab · leaked & reported = model IDs or press reports from several sources · rumor = unconfirmed. Dates are not confirmed unless the lab says so.

GPT-6.2 (Astra & Sol) release date

Leaked & reported OpenAI Expected: Mid-November (reported)

• The label “gpt-6.2-astra” has shown up in Codex usage logs, and screenshots of Google Cloud quotas listing gpt-6.2-astra and gpt-6.2-sol circulated on November 4.
• The Information reported on November 4 that new Astra, Sol and cyber models could arrive as early as November 6. The cyber model, GPT-6 Cyber, came on November 5 — but only for approved security customers. Astra and Sol had not launched by November 11; posts on X on November 7 claimed the launch was delayed over safety concerns, which OpenAI has not confirmed.
• OpenAI has not confirmed the name, a date, prices or benchmarks.

Best option today: GPT-6 Astra and GPT-6.1 Sol.

Sources: The Information, via KuCoin News · Kingy AI: GPT-6.2 Astra tracker · Fortune on GPT-6 Cyber · Posts on X about a delay (Nov 7, unconfirmed)

Next Claude Opus (Opus 5.6) release date

Rumor Anthropic Expected: Mid-November to December (rumored)

• Anthropic released Claude Fable 5.2 on November 9 — it now leads the Artificial Analysis Intelligence Index with 65 points. The newest Opus is still Claude Opus 5.5 (September 22).
• In October, users reported that some Claude Opus 5.5 requests were answered by a newer model; the name floating around for the next Opus is Opus 5.6.
• On Polymarket, “November 12” was the leading outcome for the next Opus release in early November (42%), just ahead of “no release by November 30” (38%).

Best option today: Claude Opus 5.5 and Claude Fable 5.2.

Sources: Anthropic: Claude Fable 5.2 and Mythos 5.2 · HyperDev: The next Claude rumors · Polymarket: next Claude Opus release

Qwen 4 (Max, Plus, Flash, 27B) release date

Officially teased Alibaba Expected: November (expected)

• Alibaba said on September 22 that Qwen 4 is in training. On November 6, the official Qwen account posted a teaser: “The becoming begins.”
• Four versions were named at a conference: Max, Plus, Flash and a 27B open-weight model — without specs or prices.
• Alibaba Cloud’s CTO reportedly said on October 29 that it would arrive “next week”; that week has passed.

Best option today: Qwen 3.8-Max.

Sources: Kingy AI: Qwen 4 status · Tech Insider: Qwen4 variants

DeepSeek V4.1 Pro release date

Named by DeepSeek DeepSeek Expected: No date (rumors: mid/late November)

• DeepSeek itself named V4.1 Pro on September 10, when it launched V4.1 Flash.
• Since then there has been no release date, price or benchmark; rumors point to mid or late November.
• DeepSeek’s newest model is the V4.1 Flash update from October 17.

Best option today: DeepSeek V4.1 Flash.

Sources: CometAPI: DeepSeek V4.1 Pro clues · Yotta Labs: DeepSeek V4 overview

Grok 4.8 release date

Announced by Elon Musk xAI Expected: Delayed — rumored for late November

• Elon Musk announced Grok 4.8 on September 14 as a 2.5-trillion-parameter model, to be followed by Grok 4.9 and Grok 5.
• On October 30 he wrote that it had “taken longer than expected”; there is no release date.
• Rumor lists put it between November 19 and 30.

Best option today: Grok 4.7.

Sources: Cellcog: Grok 4.8 release date · Mark Kretschmann on X (rumor list)

Tencent Hy4 open weights release date

Officially promised Tencent Expected: “In the coming days” (said Nov 4)

• Tencent released Hy4 on November 4 and said it plans to publish the model weights and system cards “in the coming days.”
• As of November 11, only the August preview (Hy4-preview) is on Hugging Face.

Best option today: Tencent Hy4.

Sources: Tencent announcement · Tencent on Hugging Face

Inkling 2 Small open weights release date

Officially promised Thinking Machines Expected: “Next week” (said Nov 5)

• Thinking Machines released Inkling 2 Small on November 5 in its Tinker platform — with closed weights for now.
• The lab said it will publish the full weights under the Apache 2.0 license “next week”, once external safety testing and red-teaming are done; as of November 11 they are not out yet.
• With 31 points on the Artificial Analysis Intelligence Index it would sit close to DeepSeek V4.1 Flash (33) and GLM-5.3 (32); the first Inkling scored 12.

Best option today: Inkling 2 Small.

Sources: Thinking Machines: Introducing Inkling 2 Small · Tinker pricing

Other dates making the rounds

These dates are forecasts only — calculated from each lab’s past release rhythm, not announced. We list them so you know what people are talking about.

Source: LLM Countdowns (forecasts, not announcements).

Just released