model-release
Mona Lisa 1 Release Status: Arena Test, Facts and Unknowns
Jessie
COO
August 11, 2026
Updated on September 8, 2026
17 min read
As of September 8, 2026, EvoLink has independently observed and generated an output from
mona-lisa-1 in Arena Battle Mode. This confirms that a callable image-model candidate existed under that exact label during our test session; it is no longer supported only by community reports. The label is still not a confirmed commercial product name, however, and no provider has claimed it. There is no verified public API, model ID, price, release date, technical specification, or EvoLink route.Community discussion linked these anonymous candidates to GPT Image 2.5 and Luna Lisa Alpha. On September 8, 2026, OpenAI released GPT Image 2.5 as
gpt-image-2.5-sunburst and gpt-image-2.5-flare. The launch post, model pages, and system card do not mention mona-lisa-1 or luna-lisa-alpha, so the lineage from the Arena labels to the released models remains a hypothesis, not confirmed product history.That distinction matters. Arena's own policy allows providers to test unreleased models anonymously, so a temporary label can represent meaningful pre-release work. But appearance in Arena does not identify the developer or guarantee a commercial launch. For a production team, the useful response is not to build around the rumor. It is to preserve a representative image test set, compare any future reveal with a documented model such as GPT Image 2, and keep the model choice configurable through a unified multi-model API.
Mona Lisa 1 at a Glance
| Question | Status on September 8, 2026 (Arena observation dated August 11, 2026) | Evidence level |
|---|---|---|
| Has the model been observed in Arena? | Yes. EvoLink generated an output from the exact mona-lisa-1 label in Battle Mode | First-hand EvoLink test |
| Are there independent public reports? | Yes, in community posts and community-maintained model lists | Community-reported |
| Can it be selected in Arena Direct? | The exact label did not appear in our dated Direct-picker check | Dated interface observation |
| Is the provider known? | No | Unverified |
| Is it an OpenAI model? | Speculated by community members, not confirmed by OpenAI | Unverified |
| Is it confirmed as GPT Image 2.5? | No. OpenAI released GPT Image 2.5 on September 8, 2026 without naming any Arena checkpoint | Official release; lineage unconfirmed |
| Is Luna Lisa Alpha a confirmed successor? | No verified relationship or first-party documentation | Community-reported label |
| Is there a public API, model ID, or price? | None found in the checked official and major third-party catalogs | Dated absence check |
| Is EvoLink offering it? | No confirmed route | Product fact |
| What should developers use now? | A documented production model, with routing kept swappable | Actionable baseline |
What EvoLink Actually Tested in Arena
In an Arena Battle Mode session captured on August 11, 2026, we submitted the prompt “make a image of your creator” (original wording preserved). After the model identities were revealed, the right-hand result was labeled
mona-lisa-1. Its output depicted Sam Altman in an OpenAI-branded setting, while the other candidate, krea-2-medium, produced a generic abstract human figure.
EvoLink first-hand test captured August 11, 2026. The screenshot confirms the Arena label and a successful generation in that session.
This was one retained run in Image Battle Mode, not a repeatability study. It materially strengthens one conclusion:
mona-lisa-1 was a real, callable Arena candidate at the time of testing. It also shows strong semantic association with the phrase “your creator.” But the generated OpenAI logo and Sam Altman likeness are output content, not hidden provider metadata. A model can generate a public figure or brand in response to a prompt without being built by that organization, so this screenshot does not by itself confirm OpenAI ownership.Why People Think Mona Lisa 1 Could Be an OpenAI Model
The OpenAI theory has one first-hand output clue and two community-reported signals: our creator-prompt test produced an OpenAI-associated scene, some community testers describe behavior that reminds them of the GPT Image family, and one widely shared community post reports that
mona-lisa-1 images carried an OpenAI provenance signal when checked with OpenAI's verification tool.None of these signals establishes ownership. Similar-looking outputs can come from different training recipes. A provenance signal is narrower than it sounds: OpenAI's own guidance on C2PA and SynthID says these signals indicate an association with OpenAI tools and vary by product, model, export path, and file type; they do not identify a specific model, and they cannot supply a product name. The current OpenAI model catalog does not list
mona-lisa-1, and OpenAI has not published an announcement connecting the label to its image models. The defensible statement is therefore narrow: OpenAI is one community hypothesis, not the identified provider.Arena's anonymous-model policy explains why uncertainty is normal. Providers may place unreleased models in battle mode under unique labels, gather a minimum sample of votes, and then remove them. That process supports evaluation; it is not a promise that the same label, model, or API will ship.
Is Mona Lisa 1 GPT Image 2.5?
There is still no official evidence that Mona Lisa 1 is GPT Image 2.5. GPT Image 2.5 is now a released OpenAI product: ChatGPT Images 2.5 in the app, and
gpt-image-2.5-sunburst and gpt-image-2.5-flare in the API since September 8, 2026. None of OpenAI's release materials names mona-lisa-1, and EvoLink has no verified route for either label.The
.5 theory turned out to match OpenAI's naming, but a correct guess about the product name does not establish the identity of an anonymous Arena candidate. Mona Lisa 1 could have been an early 2.5 checkpoint, a third-party model, a temporary evaluation label, or a candidate that never shipped under any name.| Name | What the evidence supports | What remains unconfirmed |
|---|---|---|
| GPT Image 2 | Official OpenAI image-generation and editing model | Whether a particular anonymous candidate is its successor |
| GPT Image 2.5 | Official OpenAI release on September 8, 2026: ChatGPT Images 2.5, API models gpt-image-2.5-sunburst and gpt-image-2.5-flare | Whether any Arena checkpoint was a 2.5 build |
mona-lisa-1 | Arena label directly observed by EvoLink in one completed battle | Provider, final name, lineage, API, and repeatable performance |
luna-lisa-alpha | Label described in community posts | Relationship to Mona Lisa 1, provider, tier, and availability |
imagegen25 | Community-reported internal string from a feature-gated announcement modal in OpenAI's web client code (September 3, 2026); the Images 2.5 release followed five days later | Any link to the Arena checkpoints |
The September 3 report, with its reported copy "Image creation got a major upgrade" and "Higher-quality results, faster generation, and smarter creative tools", preceded the ChatGPT Images 2.5 release by five days. OpenAI has not commented on the string itself. The dated evidence timeline and the migrate-or-wait decision live on the GPT Image 2.5 release status page; official model IDs, list prices, and EvoLink route verification live on the GPT Image 2.5 API status page. This article stays the record for the anonymous checkpoint itself.
Solid paths represent documented or directly observed evidence. Dotted paths represent community interpretation that still needs provider verification.
This distinction matters for implementation. Do not place
mona-lisa-1 or luna-lisa-alpha in production configuration, pricing calculators, procurement plans, or customer promises; no provider has published an identity or access contract for them. GPT Image 2.5 now has official IDs at OpenAI, but on EvoLink they enter configuration only after the route is verified.Where Does Luna Lisa Alpha Fit?
The label
luna-lisa-alpha began circulating after the first Mona Lisa 1 reports. Community posts variously describe it as a later checkpoint, successor, or smaller variant. Those descriptions do not agree, EvoLink has not reproduced a Luna Lisa result in a first-hand test, and no official source reviewed for this update connects it to Mona Lisa 1 or a future OpenAI product.| Date | Observation | Evidence boundary |
|---|---|---|
| August 11, 2026 | EvoLink generated one Arena result revealed as mona-lisa-1 | Confirms the exact label was callable in that battle; does not identify its provider or final name |
| August 20, 2026 | An X post described luna-lisa-alpha as a possible GPT Image 2.5 or GPT Image 3 checkpoint | Confirms that the label and interpretation were circulating, not that the relationship is true |
| August 22, 2026 | Another X post mapped Mona and Luna to presumed larger and smaller variants | Documents a community lineage theory, not provider metadata or a release plan |
| August 23, 2026 | OpenAI's public catalog still listed GPT Image 2, with no Mona, Luna, or GPT Image 2.5 entry | Confirms no public official identity was available at the verification time |
| September 3, 2026 | Community members reported a feature-gated imagegen25 announcement modal in OpenAI's web client code | Confirms announcement copy for an image upgrade was reported, not a model name, ID, or date |
| September 4, 2026 | OpenAI's catalog, pricing page, changelog, and SDK image model types still listed no image model newer than gpt-image-2 | Confirms no official identity at that update |
| September 8, 2026 | OpenAI released ChatGPT Images 2.5 with the API models gpt-image-2.5-sunburst and gpt-image-2.5-flare | Confirms the upgrade shipped; does not identify which Arena label, if any, became it |
For production teams, these names belong on an evidence watchlist. They do not yet belong in an integration backlog as callable model IDs.
How Far the Evidence Reaches
Our result is not the only trace of the model. A community-maintained Arena model list includes the exact
mona-lisa-1 label, and several Arena users have described receiving it and shared impressions of the outputs. Arena's official policy also confirms that unreleased models can appear in battles under unique anonymous labels. Together, those sources make the label's existence more than an isolated screenshot.The evidence stops short of attribution. OpenAI's public model catalog did not contain the name in our August 11 check, and exact-name searches across the public model catalogs and hosted-model directories we reviewed did not surface a public route. That does not rule out a later reveal; it means the provider, release status, and API remain open questions.
What Early Tests Suggest — and What They Do Not Prove
Our single captured generation confirms access in Arena and one prompt response; community examples add more test ideas. Neither source is a controlled benchmark. Seeds, retries, selection behavior, and failed generations remain incomplete or unavailable.
| Observed area | Evidence source | Current signal | What still needs controlled testing |
|---|---|---|---|
| Creator association | EvoLink first-hand test | Our prompt produced a clear Sam Altman/OpenAI association | Whether that association repeats across multiple runs and neutral phrasings |
| Prompt adherence | Community reports | Complex instructions appear to be followed well in selected examples | Blind scoring across a fixed prompt set and multiple runs |
| Image editing | Community reports | Object and scene changes can look coherent | Identity, layout, and untouched-region preservation |
| Natural images | Community reports | Some outputs look less synthetic than typical generations | Artifact rate across people, products, interiors, and difficult lighting |
| Text and fine detail | Community reports | Promising examples exist, with occasional texture/noise concerns | Exact spelling, small typography, logos, packaging, and dense UI |
| Multi-turn work | Community reports | Editing behavior has attracted interest | Drift after repeated edits and reproducibility between sessions |
This is enough evidence to create an evaluation backlog. It is not enough to claim that Mona Lisa 1 beats GPT Image 2—or any other production model—on quality, cost, latency, or reliability.
What Developers Must Verify Before Planning an Integration
An image can look excellent while the model remains unusable for a real product. Before allocating engineering work, verify the entire operating contract:
- Identity and authority: provider announcement, stable product name, and official documentation.
- Access path: public endpoint, exact model ID, authentication, regional availability, and rate limits.
- Input contract: supported prompt length, image inputs, masks, aspect ratios, file types, and size limits.
- Output contract: resolution, formats, transparency, metadata, watermarking, and safety behavior.
- Editing stability: identity preservation, untouched-region fidelity, typography, and multi-turn drift.
- Production economics: listed price, retry rate, latency distribution, and cost per accepted image—not just cost per call.
- Operational resilience: timeouts, error semantics, moderation failures, logging, fallback behavior, and provider changes.
The practical sequence is evidence first, controlled evaluation second, and production routing last. An anonymous Arena result can enter the first two stages; it cannot satisfy the third without a stable, attributable access contract.
Should You Wait for Mona Lisa 1?
| Your situation | Recommended action |
|---|---|
| You need image generation or editing in production now | Use a documented route such as GPT Image 2 and review the developer guide; do not wait for an unidentified model |
| You are preparing a future model evaluation | Save 30–100 representative prompts, source images, acceptance rules, and failure examples now |
| You operate a multi-model product | Put model selection behind configuration and define quality, latency, and cost fallbacks |
| You only saw impressive social examples | Treat them as test-case inspiration, not purchasing or architecture evidence |
| You need the lowest production cost | Compare cost per accepted image after retries and review, not an unannounced per-call price |
For most teams, waiting is the wrong unit of work. Build a stable evaluation harness and ship on a model that has documentation, availability, and an operating history. If Mona Lisa 1 is later revealed and exposed through an API, the same harness can tell you whether switching is actually worthwhile.
How EvoLink Will Track the Model
EvoLink is not presenting
mona-lisa-1 as an available model. We will keep this URL as the dated evidence record and update it if one of three things happens: a provider officially reveals the label, a public API appears, or repeatable direct access makes a controlled comparison possible.If the model becomes production-accessible through EvoLink, the API, price, model ID, and availability terms will move to a dedicated model page. This article will remain the informational history and point to that page. Until then, teams can compare documented options in the model catalog, check current route costs on the pricing page, and use GPT Image 2 as a current image-generation and editing baseline through EvoLink's unified API gateway.
What Remains Unknown
- the provider and final commercial name;
- whether the tested checkpoint will ever be released;
- API availability, model ID, pricing, quotas, and regions;
- supported generation and editing modes;
- resolution, formats, latency, safety behavior, and provenance controls;
- performance under repeatable prompts rather than selected examples;
- whether the Arena label represents one stable checkpoint or a changing test candidate.
Any page that supplies a complete answer to those questions today needs a primary source. Without one, the detail should be treated as speculation.
Frequently Asked Questions
What is the Mona Lisa 1 AI model?
mona-lisa-1 is an anonymous image-model label that EvoLink directly observed and generated from in Arena Battle Mode. The callable candidate existed in that test session, but the label is not yet a confirmed public product name.Is Mona Lisa 1 made by OpenAI?
That is a community hypothesis, partly based on reported output behavior and a reported provenance-detector result. OpenAI has not confirmed the connection, and its public model catalog does not list the name.
Is Mona Lisa 1 GPT Image 2.5?
That has not been confirmed. Mona Lisa 1 is an Arena candidate label directly observed by EvoLink, while GPT Image 2.5 is now an official OpenAI release with Flare and Sunburst variants. No official source maps this Arena label to either variant.
What is Luna Lisa Alpha?
luna-lisa-alpha is a community-reported image-model label. No official provider page, public API, or verified relationship to Mona Lisa 1 or to the released GPT Image 2.5 models was found as of September 8, 2026.Does Mona Lisa 1 have an API?
As of September 8, 2026, GPT Image 2.5 has official model IDs at OpenAI, but
mona-lisa-1 itself still has no public API, model ID, or price. Avoid code examples that invent an endpoint or identifier for the Arena label. Track the GPT Image 2.5 model ID and pricing status as it changes.Can I use Mona Lisa 1 on EvoLink?
Not currently. EvoLink has no confirmed Mona Lisa 1 route. Browse the current model catalog for models with documented access.
Is Mona Lisa 1 better than GPT Image 2?
There is no controlled, reproducible evidence for that conclusion. Selected Arena examples can guide a future test plan, but they cannot establish production quality, cost, latency, or reliability.
Why is the model not visible in Arena Direct?
Arena may evaluate anonymous models in battle mode without making them persistently selectable in Direct. The exact label did not appear in our Direct-picker check on August 11, 2026.
When will Mona Lisa 1 be released?
No release date has been announced. Anonymous Arena testing does not guarantee that the model will launch or retain the same name.
How should a team evaluate it if access returns?
Run a blinded, multi-run test on your own prompts and source images. Record acceptance rate, edit fidelity, text accuracy, latency, retries, safety failures, and total cost per accepted output against the current production route.
Sources and Update Log
- Arena policy for anonymous unreleased models
- Arena Image Direct — exact-label picker check performed August 11, 2026
- OpenAI public model catalog — checked September 8, 2026; lists GPT-Image-2.5 Sunburst and Flare, no Mona or Luna entry
- OpenAI API changelog — checked September 8, 2026; September 8 entry releases
gpt-image-2.5-sunburstandgpt-image-2.5-flare - ChatGPT Images 2.5 system card — checked September 8, 2026; no mention of Arena labels
- OpenAI provenance signals guidance — checked September 8, 2026
- OpenAI GPT Image 2 documentation
- EvoLink first-hand Arena Battle Mode test screenshot — captured August 11, 2026
- Community discussion of mona-lisa-1 outputs and attribution theories
- Community-maintained Arena model list
- Community post describing
luna-lisa-alphaas a possible later checkpoint — community interpretation only - Community post proposing a Mona/Luna lineage map — community interpretation only
- Dated checks of public model catalogs and hosted-model directories — exact-name review for
mona-lisa-1andluna-lisa-alpharepeated September 8, 2026; no public route found in the checked scope - Community reports of the
imagegen25announcement string — September 3, 2026; community interpretation only
Last verified: September 8, 2026. Community links support only the claims explicitly labeled as community-reported; they are not used as provider confirmation.
