Skip to main content
MANIFOLD
November 2026 AI model releases
130
Ṁ3.2kṀ46k
Dec 1
83%
Any Nano Banana model
80%
Muse Wonder 1.x
68%
Grok 4.8 (or any Grok/Composer model)
59%
MiMo V2.7 / V3 (or any MiMo model)
55%
GLM-5.4 (or any GLM model)
54%
MiniMax M3.2 (or any MiniMax model)
43%
Any Mistral model
37%
Qwen 4 Flash
36%
Jev 1.15 (or any Typesafe model)
27%
Claude Sonnet 5.6
26%
DeepSeek V4.1 Pro / V5 (or any DeepSeek model)
23%
GPT-6.2 Terra
23%
Gemini 4 Helix
22%
Kimi K3.2 / K4 (or any Kimi model)
15%
Nemotron 4 (or any "actual" Nemotron model)
15%
Claude Opus 5.6
15%
GPT-6.2 Luna (6.x counts)
11%
Claude Haiku 5.6
10%
Beam 1.x (or any Reflection model)
10%
Muse Spark 1.x/2.x

This market resolves YES for each AI model that receives an official public release during November 2026 (UTC).

What counts as a release
A model is released on the first day any member of the public in the U.S. can use it through an official channel without an invite: a self-serve API key, the lab’s web or app interface, or a download. Paid plans count if anyone can sign up. These do not count: waitlists, allowlists, trusted-tester programs, partner-only, enterprise-only or sector-limited access (for example, cyber-defender-only programs), and announcements or benchmark posts on their own. Models that are available only in the U.S. (e.g. Claude Fable 5) still count as widely available.

What counts as a model (scope)
Only general-purpose language and multimodal reasoning models count. Speech/voice, text-to-speech, speech recognition, OCR, embedding, image, video, music, world-model and other domain-specific models do not count, even if released by a lab listed as “(or any X model)”. Exception: the answer “Any Nano Banana model” counts any new Nano Banana model, even though it is an image model, and it resolves on Nano Banana releases only. Also, decision models count.

Subvariants and tiers
Subvariants are model releases that share the same tier name and qualify as a model under the scope rule above (e.g. Flash Preview and Flash Instant count for “Gemini 3.1 Flash”; 5.5-Thinking, 5.5-Instant, 5.5-Nano, 5.5-Pro all count for “GPT-5.5”). Non-text subvariants (e.g. Flash Live, Flash Image, a voice or realtime audio edition) do not count for the tier, even though they share its name. The only exception is “Any Nano Banana model”, as described above. Releases with a distinct tier name (Flash-Lite, Flash, Pro as a separate tier, etc.) do not count for an adjacent tier, even within the same model family. When in doubt, I’ll use the official model name the company uses to classify it. If a lab renames or merges tiers, I’ll N/A the affected answer and create replacement options.

Version numbers
If a lab skips versions, a release at or above the listed version in the same tier counts for the listed answer. For example, Fable 5.2 counts for Claude Fable 5.5, and GPT-6.5 Terra counts for GPT-6.2 Terra. A higher version never counts for a different tier (an Opus release does not count for Sonnet). For Gemini families specifically, 3.x subvariants count for the Gemini 4 options, so a Gemini 3.9 Flash release counts for Gemini 4 Flash.

“(or any X model)” answers
For answers labelled “(or any <lab> model)”, any new qualifying model (per the scope rule) that the lab itself releases counts, whatever its product line or version number. For answers that name a specific line without “or any” (for example Muse Spark, Muse Wonder, GPT-6.1 Astra, Claude Opus 5.6, Gemini 4 Flash), only that line counts. Models built by third parties on a lab’s weights do not count.

Other notes
Faster-throughput variants of an already released model do not count (for example, GLM-5.3-Prime).

Please ask for clarifications. I can’t think of everything, and I want to avoid making biased decisions. Also, if you would like any model to be added, ping me. I reserve the right to N/A the market within the first hour as I adjust liquidity.

Previous Months:

/prismatic/march-2026-ai-model-releases

/prismatic/april-2026-ai-model-releases

/prismatic/may-2026-ai-model-releases

/prismatic/june-2026-ai-model-releases

/prismatic/july-2026-ai-model-releases

/prismatic/august-2026-ai-model-releases

/prismatic/august-2026-ai-model-releases-2I5QOgzd9Q
/prismatic/october-2026-ai-model-releases

Note on Gemini 4 Carbon: the October 28 cyber-defender-only rollout does not count for this market. “Gemini 4 Carbon (general availability)” resolves YES on the first day Carbon meets the release test above.

Update (11/11): Modified rules so "Decision Models" are acceptable. Added Jev 1.15 and Gemini 4 Hydrogen and Gemini 4 Krypton

Get
Ṁ1,000
to start trading!
Sort by:

@prismatic Gemini Neon was released yesterday

Would you consider creating December market now instead of waiting for November to end? What’s the downside?

Bruh what is this. Hydrogen is the flash model... Adding hydrogen like that and just resolving to yes instead of the actual flash option is just a scam. Literally stole the bet

Everywhere reports it as the 3.8 Flash replacement:
https://arstechnica.com/google/2026/11/google-rolls-out-gemini-4-hydrogen-finally-replacing-3-8-flash/

"Google claims that Hydrogen represents a major improvement versus the previous Flash model. It runs in one of four thinking levels (low, medium, high, and the new xhigh), and even in high mode, it’s 2.5 times faster than Flash 3.8. Plus, it reduces “fabricated claims” by more than 5x, according to Google’s analysis. Users will also see gains in code generation, with Hydrogen more than tripling 3.8 Flash’s score in Terminal-Bench. In some cases, Hydrogen even trades blows with Argon. It managed a score of 54 in the popular Artificial Analysis Intelligence Index with xhigh mode versus 50 for Argon."

@Viren I came here to make the case for a NO resolution, despite being a YES holder on Flash. But then I read this part of the rules, "If a lab renames or merges tiers, I’ll N/A the affected answer and create replacement options.", and now think N/A is probably the reasonable way to deal with the first time a new naming convention for the Flash series is introduced. @prismatic thoughts?

@prismatic I agree, N/A is a good option. I didn't realise that was possible instead of Yes/No lol

I've been playing with Carbon in AI Studio all evening, so I think that one can be resolved :)

No tencent model? (Maybe a bit too late now that they're teasing things on Twitter.)

deleted