Anthropic may be preparing not one, but two new Claude models for release.
Two mysterious identifiers have surfaced in developer circles: claude-melon-eap and claude-marshmallow-eap. They are reportedly early-access test versions of upcoming Claude models, although Anthropic has not officially confirmed either identifier or what products they will eventually become.
The speculation is getting particularly interesting because some users reportedly appear to be experiencing changes in how Claude routes requests, raising the possibility that Anthropic is already testing newer model checkpoints behind the scenes.
If the reports are accurate, Anthropic could be preparing a double Claude release.
Two New Claude Identifiers Have Appeared
The two identifiers attracting attention are:
claude-melon-eapclaude-marshmallow-eap
The eap suffix is widely interpreted by the developer community as referring to an Early Access Program or preview build. However, Anthropic has not publicly defined these identifiers or confirmed that they represent final consumer-facing models.
That distinction is important.
A leaked model identifier doesn’t necessarily mean a finished product is ready for launch. Internal checkpoints can be renamed, merged, changed or abandoned before they ever reach the public.
Still, the appearance of both names has sparked plenty of speculation about what Anthropic might be testing.
Is Melon Fable 5.1?
One of the leading rumors currently links claude-melon-eap with Fable 5.1.
If that turns out to be correct, Anthropic could be preparing a relatively quick upgrade to its highest-end Claude model.
Fable 5 is already positioned as Anthropic’s highest-capability model for demanding, long-running agentic work. A Fable 5.1 release would therefore be particularly interesting if Anthropic has managed to improve reasoning, reliability, tool use or long-horizon performance without requiring an entirely new generation.
However, the Melon → Fable 5.1 connection remains unconfirmed. Other reports have proposed different mappings, including Sonnet-class models.
Marshmallow Is Reportedly Opus 5.1
The second identifier, claude-marshmallow-eap, has been more closely associated with a possible Claude Opus 5.1.
Some early reports claim Marshmallow performs better than the current Opus 5 in certain tasks, while other reports describe it as more pleasant or consistent to use.
That has led to the theory that Anthropic could be preparing an updated Opus model that addresses some of the complaints surrounding Opus 5.
But again, this is speculation. Anthropic has not confirmed that Marshmallow is Opus 5.1, and there is currently no official Opus 5.1 model card, public API identifier or release announcement.
Some Users May Already Be Seeing Something New
Perhaps the most intriguing part of the story is what users are reportedly seeing in Claude.
Some Claude users have claimed that their experience with Opus 5 has changed, with the model appearing sharper or behaving differently. This has prompted speculation that Anthropic may be silently routing a portion of users toward a newer checkpoint while keeping the visible model name unchanged.
That’s possible, but it hasn’t been proven.
A different system prompt, routing configuration, reasoning settings or A/B test could produce similar changes without requiring a new model to be deployed.
In other words, a better-performing Opus 5 session doesn’t automatically mean you’re secretly using Opus 5.1.
Recent analysis of the leak specifically notes that users reported these changes while the interface continued to identify the model as Opus 5. No public backend identifier has confirmed that a 5.1 checkpoint is being served.
Why a Double Release Would Make Sense
A simultaneous or closely timed release of two models would fit Anthropic’s increasingly aggressive model development cycle.
The company released Fable 5 in June, Sonnet 5 later that month, and Opus 5 in July. That means Anthropic has already been moving quickly across its model lineup in 2026.
If Melon and Marshmallow are genuine next-generation checkpoints, Anthropic could be preparing another rapid refresh rather than waiting months for a major numbered generation.
That could mean something like:
Melon → Fable 5.1
Marshmallow → Opus 5.1
But it’s worth emphasizing that this is the current rumor, not an official Anthropic product roadmap.
Anthropic May Be Testing More Than Just Model Intelligence
The -eap identifiers are also interesting because they suggest that Anthropic could be testing these models with a limited audience before a wider rollout.
That would allow the company to evaluate real-world behavior, latency, safety systems, routing and user feedback before deciding whether a checkpoint is ready for everyone.
There are also reports of a related identifier, claude-marshmallow-ht-eap, appearing in API traffic. One investigation reportedly found fallback behavior to an older Opus model during some safety-sensitive requests. The evidence around these identifiers remains incomplete, but it suggests that the testing may involve more than simply evaluating raw benchmark scores.
The Timing Is What Makes This Interesting
Anthropic doesn’t appear to have confirmed a release date for either rumored model.
But the timing has created plenty of speculation.
If Anthropic really is preparing Fable 5.1 and Opus 5.1, the company could be looking to keep pressure on OpenAI and other frontier AI labs rather than allowing competitors to establish the next benchmark lead.
And with OpenAI also reportedly working on its next-generation Astra system, the AI model race is becoming increasingly competitive.
A new Claude release right now could make the next few weeks particularly interesting.
What We Know vs. What We Don’t
Here’s the important distinction:
What appears credible:
- Two identifiers,
claude-melon-eapandclaude-marshmallow-eap, have circulated in developer communities. - The identifiers appear to be associated with unreleased or early-access testing.
- Developers have been investigating their behavior and possible model positioning.
- Some users have reported changes in their Claude experience.
What remains unconfirmed:
- Melon is Fable 5.1.
- Marshmallow is Opus 5.1.
- Either model will actually launch.
- Anthropic is secretly routing users to a 5.1 model.
- A public release date exists.
- The two models will launch together.
Those details could change quickly if Anthropic makes an announcement.
Anthropic Could Have a Big Surprise Coming
For now, Marshmallow and Melon are just names attached to a potentially much bigger story.
If the current theories are correct, Anthropic could be preparing two upgraded Claude models at the same time, with one targeting the Fable line and the other potentially becoming Opus 5.1.
And if some users are already being routed toward newer checkpoints, Anthropic may be further along than the public evidence suggests.
Nothing is official yet.
But if Anthropic really is preparing a double Claude release, the next model battle between Claude and its biggest AI rivals could get very interesting, very quickly.
For continuing updates on OpenAI, ChatGPT, Codex, GPT models, and AI industry developments, bookmark:
