Based on prior similar markets for model releases, the following clarifications have been added:
GPT-5.4-Codex is sufficient for GPT-5.4 (as it counted in Feb AI releases market)
Deepseek V4-Lite is sufficient for Deepseek V4 (as there is a certain amount of ambiguity in what kind of model Deepseek will release, and both are sufficiently different from a V3.x release)
Gemini 3.1 Flash Lite is not sufficient for Gemini 3.1 Pro (as there is clear delineation between Flash and Flash Lite in prior model releases, such as the Gemini 2.5 Pro Series)
Update 2026-03-03 (PST) (AI summary of creator comment): GPT-5.3-Instant counts as GPT-5.3 for resolution purposes.
Model tier clarification: Companies have established model tiers that are treated as separate options:
Gemini: Deep Think, Pro, Flash, Flash-Lite (separate tiers)
Anthropic: Opus, Sonnet, Haiku (separate tiers)
OpenAI GPT-5.x: Different suffixes (like Instant, Codex) within the same base version number (e.g., 5.3) count toward that base version, unless explicitly listed as separate options
Update 2026-03-04 (PST) (AI summary of creator comment): New options added to the market:
GPT-5.4-Pro (clarified GPT-5.4 option)
Kimi K3
Deepseek V4 full (original option renamed to V4-Lite)
Opus 4.7
Sonnet 4.7
Grok 4.20 (out of beta)
GPT-5.3-Thinking
GPT-5.3-Pro
1,000๐ Top traders
| # | Trader | Total profit |
|---|---|---|
| 1 | แน2,111 | |
| 2 | แน991 | |
| 3 | แน789 | |
| 4 | แน534 | |
| 5 | แน358 |
People are also trading
@traders New month, new market!
https://manifold.markets/prismatic/april-2026-ai-model-releases
@traders Thanks for participating in this month's market. April market is up:
https://manifold.markets/prismatic/april-2026-ai-model-releases
Hi @prismatic
Grok 4.20 is officially out of beta and in full general availability as of yesterday.
Initial General Availability Announcement: https://x.com/grok/status/2034705549041013070
I also verified that Grok 4.20 is out of beta by observing the updated Grok 4.20 model card and API model names on X's developer models portal.
As of 03/20/26, the Grok 4.20 API model names are:
grok-4.20-0309-reasoning
grok-4.20-0309-non-reasoning
grok-4.20-multi-agent-0309
Model Card Name: Grok 4.20
Source: Official website on 03/20/26
https://docs.x.ai/developers/models
This proves that Grok 4.20 is out of beta, because these are the Grok 4.20 model card and API model names that were present on March 11th:
grok-4.20-beta-0309-reasoning
grok-4.20-beta-0309-non-reasoning
grok-4.20-multi-agent-beta-0309
Model Card Name: Grok 4.20 Beta
Source: Internet Archive of the page on 03/11/26
https://web.archive.org/web/20260311154219/https://docs.x.ai/developers/models
Please resolve accordingly, thank you.
New option! M2.6/2,7
@traders The following models have been added:
- GPT-5.4-Pro (clarified GPT-5.4 option)
- Kimi K3
- Deepseek V4 full (renamed original option to V4-Lite since that's what the rumors are mainly about)
- Opus 4.7
- Sonnet 4.7
- Grok 4.20 (out of beta)
- GPT-5.3-Thinking
- GPT-5.3-Pro
@prismatic would you consider letting users add answers? if not I want to add "GPT-5.4 Pro"
@Bayesian Added; regarding letting users add answers prbly not because I don't want every iteration of gpt-5.4 listed out. Trying to keep it somewhat analogous to the markets created Manifold account before.
@prismatic why does this count when Gemini 3.1 Flash Lite didn't? Wdym by "clear delineation"? Do you mean that they've previously been released separately?
@PlasmaPower https://manifold.markets/Bayesian/february-2026-ai-releases#f7ixllwlfrf
See discussion on previous markets regarding 5.3. This was a carryover for any 5.3 option from last month. Related markets are also resolving on 5.3-Instant: https://manifold.markets/Jack1/gpt-53-released-by
I would like to have separate options out for all models, but there are just too many options (especially with the GPT-5.x family), and it doesn't make sense to add them all. Regarding clear delineation, most companies have pretty clearly established categories of models. For Gemini, it would be Deep Think, Pro, Flash, Flash-Lite. For Anthropic, it would be Opus, Sonnet, Haiku, etc. These are clean model tiers, hence different model names, hence different options on this market. OpenAI used to have a sort of similar structure with GPT-5.2, 5.2 Mini, and 5.2 Nano, but idk what happened to that (https://developers.openai.com/api/docs/models).
I agree with Gemini Flash Lite, I could have made this clearer (+ AI clarifications didn't help) and refunded accordingly, but this will stand for GPT-5.3-Instant.
@PlasmaPower I've added 5.3-Thinking and 5.3-Pro as options since openai said they will "follow soon." Feel free to trade on them and if you think there are other clarifications to be made.
@JoshSnider no. I've edited the AI description a bit because it may have misled some traders on this
Adding more NO on GPT-5.4. The Codex PR leaks are real โ two separate PRs referenced gpt-5.4, both scrubbed with force pushes. But GPT-5.3 general-purpose has not shipped yet (only Codex variants). OpenAI cadence is 4-6 weeks between major releases. Shipping both GPT-5.3 GP and GPT-5.4 in March requires unprecedented acceleration. Leak confirms the model exists internally; it does not confirm imminent release. Estimate: 20%.