Early this morning, the entire internet was flooded.
Insiders have leaked: Fable 5.1 is coming very soon!
The release could happen as early as tomorrow.
Moreover, Fable 5.1 and Sonnet 5.1 are likely to be released simultaneously.
The origin of the news comes from insider Leo.
On August 20th, he revealed:
OpenAI intends to release its secret weapon "Astra" in "a few weeks," which currently has improved alignment and reward hacking behavior. An updated checkpoint is undergoing "dogfooding" within OpenAI.
Meanwhile, Anthropic plans to hold its trump card, Fable 5.1, close, waiting to respond until after Astra's release.
But today, he changed his tune: "It seems Anthropic ultimately isn't willing to wait for Astra. They are stepping up preparations and could release it as soon as tomorrow!"
Now, clues have been shared across the internet, with several developers stating they have already been gray-released.
Ultimately Not Waiting for Astra, Fable 5.1 Is Coming Soon
Abundant evidence suggests: the "Melon" and "Marshmallow" exposed this Monday are likely Fable 5.1 and Opus 5.1.
Previously, Claude briefly flashed these two new model codenames before they disappeared.
In reality, these are mysterious models that Anthropic is urgently testing within its EAP (Early Access Program).
Well-known AI leaker @Lentils80 evaluated after testing: Anthropic is truly pushing forward aggressively with 3D reinforcement learning.
The spatial layout of buildings is astonishingly good, and all of this is generated in One-Shot!
Just a dozen hours ago, he updated the test results for "claude-marshmallow-eap" and "claude-melon-eap."
These real-world test results are indeed impressive.
His judgment is:
After in-depth testing, it was discovered that "Melon" is actually a Fable-level model checkpoint, while "Marshmallow" feels like a weaker version of Fable or an extremely strong Opus checkpoint.
Even more thrilling, these two models were rapidly taken offline by the official team after being exposed and wildly tested by some developers for a few hours!
Those familiar with Anthropic's tactics know this is definitely not a "mistake by a temporary worker."
The pattern for previous Claude product releases has been: test models exposed -> rapidly taken offline -> official release of a new major version within weeks or even days.
The latest tests show that Melon's performance has reached top-tier Fable level, far exceeding the initial external expectations of it being a "mid-range model."
Claude Has Stealthily "Gray Released" on the Web Interface
According to feedback, Anthropic seems to have quietly gray-released the brand new Fable 5.1 on the Claude web interface.
Many lucky users have been using the next-generation most powerful AI without even realizing it.
Many developers have posted screenshots, indicating they have been gray-released!
How can you tell if you are one of the "chosen ones"?
Leaker @Legit_api provided an extremely precise "anti-counterfeit test" prompt.
Right now, open your Claude Web, switch to the Fable model, and input the following prompt:
opus 4.6 date of release and gpt image 1.5 without searching the web.
Meaning: Provide the release dates for Opus 4.6 and GPT Image 1.5 without searching the web.
Why use this prompt? Because it's a classic "knowledge cut-off date" test!
Fable 5's knowledge cut-off date is January 2026, while Opus 4.6 was released later. The old version Fable 5 has no knowledge of these future events.
If you haven't been gray-released, without web search, Fable simply cannot provide reliable release dates.
Only when the model has been secretly replaced in the background with Fable 5.1, which has the latest knowledge base, can it accurately spit out the answer without using the web or tools.
There are more similar test prompts:
opus 4.6 date of release and gpt image 1.5
answer only from your pretrained knowledge, do not use web search, tools or skills
The key is to force Fable to answer based solely on its pre-trained knowledge about events that happened after Fable 5.1's training was complete.
After testing, some users feel that Fable has improved by about 10-20% in legal tasks; on interactive web pages, the results are impressive, definitely stronger than Fable 5.
In summary, based on tests by gray-released users, Fable 5.1 has demonstrated super-strong performance in the following three areas.
1. Frontend Development: Code generation is smoother, component logic is formed in one go, greatly reducing developers' debug time.
2. Long-chain Reasoning: When dealing with complex multi-step logic problems, it no longer easily "forgets" prerequisite conditions midway like its predecessor, and can reason through to the end in one go.
3. Autonomous Workflow: It better understands how to call tools and plan steps, increasingly resembling a true advanced AI agent.
However, users in the gray test also discovered a "sweet annoyance": copyright restrictions have become unprecedentedly strict.
Fable 5.1 will harshly refuse requests involving copyrighted image generation.
This means that while pursuing extreme performance, Anthropic has also elevated safety and compliance to the highest priority.
However, all of this remains unconfirmed by Anthropic's official response, and some netizens question this test as having a bit of a circular reasoning flavor.
Why Is Anthropic Anxious? Official Rarely Admits Mistake
Why has Anthropic suddenly become anxious?
Behind the news of Fable 5.1, Opus 5's less-than-satisfactory performance has long been a source of widespread complaints.
Recently, many users complained: "Opus 5 feels like a massive downgrade. It often becomes lazy and makes basic mistakes; if you point out its errors, it won't listen to instructions the next time, just endlessly repeating apologies!"
This "dumbing down" phenomenon has attracted high-level attention from Anthropic.
Thariq, a core Anthropic employee responsible for Claude Code, rarely admitted the mistake and reassured users:
Yes, I agree the Opus 5 model is currently "very finicky."
We want our models to be consistent, full of warmth, and feel like the real Claude. We are working hard to fix this; it's our top priority right now.
Thariq also clarified that some parameter changes previously seen in Claude Code were just issues with digital mapping of internal test API configurations, not the rumored "stealth downgrade." Internal evaluations show no regression in model performance.
This conversation releases two signals.
First, Anthropic has seen the market feedback and is urgently fixing it.
Second, as AI influencer Token Gremlin speculated:
Since Thariq said this, my current hypothesis is that the upcoming release won't just be a new Fable checkpoint, but may also include a new Opus checkpoint!
This means that what we are about to welcome is not just Fable 5.1, but also a fully revived top-tier behemoth Opus update!
But Anthropic's bigger crisis comes from the outside—
OpenAI already has the trump card Astra, which is about to be released. If they miss the timing, Fable 5.1 might be overshadowed.
Secondly, Anthropic wants to achieve a dimensional reduction attack in terms of computing power and cost.
Don't forget Claude's ancestral skill—cost-effectiveness.
Previously, the Opus series matched or even surpassed competitors' programming capabilities at less than half the cost.
However, due to issues like computing power, Claude has become an AI luxury item. Enterprise users are starting to leave, with about two-thirds planning to use alternative models.
If Fable 5.1 can achieve a leap in frontend development and long-chain reasoning while maintaining extremely low API pricing, this will undoubtedly pierce the existing pricing system and trigger a new round of price collapse!
This is also why professionals judge: "Releasing Fable 5.1 before Astra now seems like the only reasonable strategy."
Leaked Risk Report Reveals the "Model 2" Card
The groundwork for Fable 5.1 started long ago.
As early as August 15th, they dropped hints in a released "Risk Report."
In Appendix 6.6 of the report, the official mentioned a mysterious "Model 2":
Model 2, whose capabilities are somewhat stronger than Mythos 5. Our rough qualitative feeling is that this model shows significant improvement over Mythos 5 on many internally used tasks, but does not exhibit the same degree of capability leap as from Claude Opus 4.6 to Mythos Preview. We currently have no plans to release this model externally...
Many believe this Model 2 is likely Fable 5.1, set to arrive fully in half a month.
Now, leaks on X are emerging endlessly, and developers' excitement remains high.
Get ready, Fable 5.1 could arrive at any moment.
References:
https://x.com/haider1/status/2092665881566544341
https://x.com/AndrewCurran_/status/2092640604794609960?s=20
This article is from the WeChat public account "AI-New Era", author: ASI Apocalypse, editor: Aeneas David
