Anthropic has delayed the release of its next-generation large model, Claude Opus 5, due to security concerns, sparking widespread attention. Although the model, codenamed “Hive,” features groundbreaking capabilities such as a 1-million-token context length and configurable reasoning intensity, and has demonstrated exceptional performance in 3D detail generation and code writing during testing, the company has postponed its launch citing potential threats to human control over advanced AI. Netizens have expressed strong dissatisfaction, accusing Anthropic of disguising product delays as safety concerns and questioning its “hunger marketing” strategy. The model has also been found to suffer from a significant weakness in basic visual recognition.
Author and source: AIERA
Watched the entire network all day, ended up with nothing.
Throughout Thursday, the entire network was watching for the rumored Opus 5, aside from keeping an eye on the Phi Award.
After all, the atmosphere has already been set.
Rumors from insiders, leaked screenshots circulating widely, and numerous signs all indicate: Tonight, Opus 5 will surely arrive.
As a result, everyone waited all day, but nothing happened.
Just now, a well-informed source said: Don’t wait—Opus 5 has been put on hold.
The reason remains the same: they fear this model could threaten human control over advanced AI. One insider even called it the first model to make the lab question whether the Scaling Law should continue.
As a result, the entire network exploded.
Netizens angrily criticize: Stop the hype already.
What threatens humanity, what questions the Scaling Law—such claims might have fooled the public two years ago, but after countless “boy who cried wolf” episodes, netizens are no longer buying it.
It’s clearly a delay, yet you’re packaging it as “for the safety of all humanity”? A/, you’re truly still the same A/.
“Stop using these mindless scare tactics to frighten people!” someone angrily snapped: “This kind of panic-mongering is annoying—please just stop, thank you.”
More netizens believe this is simply a hunger marketing tactic by Anthropic when its product fell short of expectations.
Anthropic has a condition called compulsive lying.
Some also say that Anthropic’s continuous efforts to create panic and confusion have already driven people to switch to other models—the game is over.
Early test preview: 3D details are legendary
So, how does the legendary Opus 5 actually perform?
Some developers with quick access and special permissions have already begun testing Opus 5 using certain methods (rumors suggest that some Opus 4.8 requests are being automatically routed to Opus 5).
Popular influencer Chetaslua released multiple demos, enthusiastically stating: “Opus 5 is a 3D beast!”
He had Opus 5 generate a highly complex weather UI card in one go, with a level of detail that is impressive.
The game generated by Opus 5 made him say his hands were shaking.
I swear I’ve never seen such intricate detail all at once in my life—not even multiple rapid shots could capture this effect; my hands are shaking from excitement. Oh my goodness, this piece is incredible.
Regarding this 3D home generated by Opus 5, he commented: “This is AGI.”
This advantage in spatial understanding, three-dimensional detail construction, and UI code generation demonstrates that Anthropic has indeed made significant breakthroughs in foundational logical reasoning and code generation capabilities.
However, the cost of acquiring this ability is extremely high.
Chetaslua revealed: Opus 5 is incredibly token-intensive! After just three conversations, all my quota was exhausted—even Fable 5 doesn’t consume tokens this heavily.
Additionally, in real-world testing, Opus 5 also performed poorly on basic image recognition tests.
On June 23, he discovered during testing that Fable 5 failed a visual test by failing to identify a small insect on a pancake in the image, while Google’s model handled it easily.
Today, he tested Opus 5 with the same questions and found it still failed.
Perhaps Opus 5 really does have serious imbalances.
Opus 5, codenamed Hive
In fact, everyone had high expectations for Opus 5 previously.
The reason we believe Opus 5 is coming soon is primarily due to clues from Cursor.
According to leaks, Opus 5 is a well-rounded all-rounder with no weaknesses.
A prominent influencer once predicted that, if the leaked information is accurate, this would be Anthropic’s greatest release in a long time.
As originally planned, Opus 5 will bring a transformative leap:
Exceptional memory: Supports an ultra-long context window of up to 1 million tokens, allowing you to feed it entire libraries of novels or massive codebases for analysis.Configurable inference effort: Users can dynamically adjust between speed and quality, according to their preferences.Full Evolution: Significantly enhanced in coding, acting as an AI agent, using tools, and multi-step reasoning.Smart Security and Fallback Mechanisms: Provide smarter security protection for production workflows.
Independent developer Pankaj Kumar also revealed that Opus 5’s capabilities will come incredibly close to those of the top-tier Fable 5, while costing half the price, making it a highly sought-after budget alternative for developers.
On July 9, chetaslua discovered a mysterious option named “Claude Honeycomb EAP” in the model selector of Cursor IDE.
The model description codenamed “Honeycomb” appears highly advanced: an Anthropic research model featuring per-turn safety controls and safety fallbacks, a 1 million token context window, and extremely high reasoning capacity.
Although this option was quickly removed after only two prompts were processed, the community immediately realized: this hive might be the rumored Opus 5!
Notably, the “Hive” architecture is ingeniously designed with a reverse security fallback mechanism.
When the model encounters a complex issue and determines its confidence level is insufficient, it does not fabricate answers like other large models; instead, it triggers a fallback mechanism, automatically downgrading to invoke Opus 4.8 as a backup.
This design skillfully balances computational cost with output stability, affirming its ecosystem positioning: more powerful than Opus 4.8, yet more affordable than Fable 5.
Originally, this was supposed to be Anthropic’s new flagship offering to the market.
Why is Anthropic rushing to reveal Opus 5 at this time?
The answer is simple—it’s been incredibly competitive lately!
Old rival OpenAI is closing in, while Tibo is resetting quotas as if they cost nothing, causing countless developers to check first thing in the morning whether “Saint Tibo” has announced a reset yet.
On the other hand, Kimi, Qwen, and GLM are steadily gaining ground in winning the hearts of developers worldwide.
Opus 5, codenamed “Hive,” was originally tasked with a clear mission: to overcome the high computational barriers using top-tier reasoning capabilities, and to demonstrate what a cutting-edge model looks like with 1 million context tokens and 3D generation capabilities.
But now, Anthropic really doesn’t have much time left to drip-feed updates.