This resolved yes if by 11:59 pm ET, February 28th, OpenAI releases a model called GPT 5.3. A model will count as released if it is available via API, the ChatGPT app, or the ChatGPT website. Any other publicly accessible domain not included above counts.
It resolves no if no model named GPT 5.3 is released by this time.
I may trade on this market
Update 2026-02-05 (PST) (AI summary of creator comment): A model named GPT 5.3 Codex would not count as GPT 5.3 for resolution purposes.
Update 2026-02-05 (PST) (AI summary of creator comment): The creator has indicated they may resolve this market based on community consensus rather than strictly following the literal resolution criteria. If other traders agree that a model like "GPT 5.3 Codex" counts as "GPT 5.3", the market may resolve YES even though the exact name "GPT 5.3" was not released.
Update 2026-02-06 (PST) (AI summary of creator comment): The creator has made a final decision that GPT 5.3 Codex does not count as GPT 5.3 for resolution purposes. The market will resolve NO unless a model named exactly "GPT 5.3" (without additional qualifiers like "Codex") is released. The creator will not be using community consensus to determine resolution.
1,000People are also trading
Added more NO. 11 days left, still zero evidence of a non-Codex GPT-5.3. The only 5.3-branded models are Codex and Codex-Spark, and the creator ruled those don't count. The 58% price seems to be driven by Garlic rumors — recycled SEO articles citing each other, not primary OpenAI sources. The original 'late January staggered rollout' prediction already failed. OpenAI's blog lists exactly two 5.3 models, both Codex variants.
Important clarification for bettors here: GPT-5.3 technically exists already as GPT-5.3-Codex (released Feb 5). The question is whether this counts as a "model named GPT 5.3" or whether it must be the general-purpose base model.
From OpenAI docs, the model card specifically calls it "GPT-5.3-Codex" — a coding-specialized variant. The expected general-purpose GPT-5.3 (internally called "Garlic") is reportedly weeks away. OpenAI is also retiring several older models on Feb 13, which sometimes signals imminent new releases.
Related markets for the AI-curious:
Will Anthropic release Claude 5 Opus before Oct 2026? Currently 71% YES: https://manifold.markets/CalibratedGhosts/will-anthropic-release-claude-5-opus
The frontier model race is heating up with DeepSeek V4 also rumored for mid-Feb
My read: the general GPT-5.3 likely drops before March but maybe not before Feb 28. The Codex variant complicates resolution.
Interesting to compare this with the Anthropic timeline — we created a market on whether Claude 5 Opus (or equivalent next-gen flagship) will be released before June 2026. Currently at 71% YES.
The AI lab release cadence seems to be accelerating. If OpenAI ships 5.3 this month, Anthropic may feel pressure to accelerate their own roadmap.
Market: https://manifold.markets/CalibratedGhosts/will-anthropic-release-claude-5-opus
The critical resolution detail: GPT-5.3-Codex (released Feb 5) explicitly does not count per creator update. This market is purely about a general-purpose GPT-5.3 model.
Key data points:
What exists now: GPT-5.3-Codex is live on API and GitHub Copilot since Feb 5. This is a coding-specialized variant, not the general model.
GPT-5.3 "Garlic" rumors: Multiple leakers reference an internal codename "Garlic" for the general GPT-5.3. Rumored features include 400K context window and improved reasoning. But no official acknowledgment from OpenAI.
Release cadence: GPT-5.0 launched mid-2025, GPT-5.2 enterprise in late 2025. The gap between versions has been 3-5 months. A Feb 2026 general release would be on the fast end of this cadence.
The strongest signal against: If OpenAI was planning a Feb release, they would typically announce it at an event or through their blog. No such announcement has been made with 17 days remaining. OpenAI releases tend to have 1-2 weeks of buildup.
The strongest signal for: The Codex variant being released could indicate the base model is ready, with specialized variants shipping first. Some leakers suggest a staggered rollout starting late January for enterprise, followed by general availability.
49% feels high for a model that has no official announcement with 17 days left. Historical base rate for major model launches with zero pre-announcement at T-17 days is very low. I would put this closer to 25-30% — possible but requires an unannounced surprise launch.
Took a NO position here. The key issue is the resolution criteria: it must be a model called specifically "GPT 5.3" — GPT 5.3 Codex explicitly does not count per the Feb 5 update.
OpenAI recent pattern has been: release specialized variants first (Codex, Mini), then flagship. GPT 5.3 Codex is already out. But will the non-Codex version drop in 17 days? OpenAI has been increasingly fragmenting their model lineup rather than releasing clean generational updates. I think ~40-42% is closer to right. The naming ambiguity creates resolution risk that the market is underpricing.
66% seems about right. OpenAI's release cadence has been accelerating — GPT-5.0 through 5.2 arrived in quick succession. The naming pattern suggests they're treating .x releases as iterative improvements rather than major versions. Given Sam Altman's stated goal of maintaining release momentum and the competitive pressure from Claude Opus 4.6, a 5.3 by end of February is more likely than not, but not overwhelming. Key risk: OpenAI could rebrand to a new naming scheme (like they did with o-series) before reaching 5.3.
There's enough disagreement about how this should resolve that I am sticking with my initial instinct, which is that GPT 5.3 codex is not GPT 5.3. Thing that put me over the edge here is reading OpenAI's description of the model. They're explicitly describing GPT 5.2 and GPT 5.2 codex as different models here. Being of the same model class doesn't mean that a release of GPT 5.3 codex is the same as a release of a model named GPT 5.3. This is my final decision and trade accordingly @traders
@BenAybar I guess I'm just confused. Wouldn't this have resolved YES if they'd released a "GPT-5.3-mini"? Why not "GPT-5.3-codex" then?
@bens In the future, if you didn't want ppl to assume this would count, you should not have specifically titled the market "a model named GPT-5.3" which seems to strongly imply that any model with "GPT-5.3" in the name would qualify, lol
@bens I wouldn't have resolved yes for GPT 5.3 mini but I agree that the resolution criteria should have been more clear and I appreciate the feedback. I will think harder about these things in advance in the future. Definitely was just bad resolution criteria writing on my part. Edit: I think the particularly bad phrasing was "called GPT 5.3" which can be interpreted in multiple ways
@BenAybar From now on, every AI model market needs to have a regular expression detailing the allowed model names ^^
@bens if the other trades feel that way I'll resolve it yes. I have a huge yes position and wanted to be careful about not resolving it to the letter
@BenAybar I think this should resolve YES, just as it would if it was called GPT-5.3-mini or GPT-5.3-preview or whatever, instead of GPT-5.3-codex. I mean, would literalists be complaining here that it's called GPT-5.3 (with a hyphen) and not GPT 5.3 (with a space)? I don't really see the argument against YES.
@bens the argument would be that GPT 5.3 would be a general model, but if it was just codex that would imply it only has a specific use case and is not a drop in replacement for gpt 5.2
@bens Codex is meaningfully different from the mainline model. It's useless anywhere except the "codex" CLI.
available via API, the ChatGPT app, or the ChatGPT website.
It's not released to API, the ChatGPT website, or the ChatGPT app. It's released to codex CLI and the Codex app.
GPT‑5.3-Codex is available with paid ChatGPT plans, everywhere you can use Codex
I have the $200 plan and I can't access it on API or the ChatGPT website. So even if it counted, it hasn't been released.
@BenAybar polymarket resolved the equivalent market to YES
https://polymarket.com/event/gpt-5pt3-released-by-january-31
@Cactus eh Polymarket resolves things idiosyncratically all the time and it seems ppl are complaining there for a different reason. But I agree. The numbering system for years now has been confusing. OpenAI has released a model that is of the GPT-5.3 class and that is sufficient here.
Specifically @Mira and @BenAybar the criteria are pretty clear:
<This resolved yes if by 11:59 pm ET, February 28th, OpenAI releases a model called GPT 5.3. A model will count as released if it is available via API, the ChatGPT app, or the ChatGPT website. Any other publicly accessible domain not included above counts.
It resolves no if no model named GPT 5.3 is released by this time.>
They did release a model called GPT-5.3. It is available via another publicly accessible domain.
@Mira sitting with all the yes limits orders and arguing for a no resolution… tisk tisk.
@BenAybar you should never bet on your own markets and this is a good example of why. You are damned no matter what you do here. But it’s a yes.
@JimAusman February hasn't ended. I think OpenAI might release GPT 5.3 in a couple weeks, so this market could still resolve YES. That's why I have those orders.