Just Now, Mythos 5 Cracked a Major Quantum Computing Challenge

marsbitPublished on 2026-10-17Last updated on 2026-10-17

Abstract

Anthropic's Mythos 5 has achieved a breakthrough in quantum computing by autonomously designing quantum circuits that outperform human-optimized designs. Given only plain-language descriptions of eight foundational quantum circuits in chemistry, optimization, physics simulation, and number theory, the AI ran a fully automated closed loop: it wrote circuit designs in Python, received scores based on qubit count and gate count, and iteratively refined its work without human intervention. Within 53 hours, it beat the best publicly available human designs on all eight tasks. Two results stand out. First, in modular inversion for elliptic curve cryptography, existing divsteps algorithms were considered impractical on quantum computers because they generate branch history that cannot be discarded due to reversibility requirements. Mythos 5 derived entirely new algebraic identities to reconstruct that history from the result, producing the first reversible quantum divsteps implementation that needs no branch storage—cutting gate count by 17x with only 12% more qubits. Second, it redesigned the Reed-Solomon decoder inside Google's Decoding Quantum Interferometry algorithm, replacing the extended Euclidean approach with Berlekamp-Massey, which matches quantum circuit constraints. It then used a self-written simulator running 403 million samples to measure exception rates and reduced memory from 665 qubits to 19, achieving a 12x reduction in gate count. The work suggests software can...

Just moments ago, Anthropic unveiled a bombshell study.

Mythos 5 has autonomously closed the loop, designing entirely new quantum circuits far more efficient than humanity's "best solutions"!

And it did so from a purely natural-language description.

In a key decoding task, Claude directly slashed computational cost by 12x, even 17x!

Even more astonishingly, to complete the task, it autonomously derived entirely new algebraic identities that humans had never discovered.

Within 53 hours, Mythos 5 crushed years of human research.

One netizen exclaimed: "Every physics PhD just realized the paper they've slaved over for years—AI can write a version 10x better with just a few plain sentences..."

Humanity Was Stuck: The "Valley of Despair" in Quantum Computing

Why is designing quantum circuits so hard?

You see, traditional computers use 0s and 1s (bits) to store information.

Quantum computers use "qubits." Leveraging quantum superposition, they can process vast amounts of information simultaneously. Once truly practical, they could crack encryption, develop cures for terminal diseases, discover room-temperature superconductors...

However, current quantum computers have major flaws.

First, there are too few qubits. Today's machines max out at a few hundred to a thousand qubits—nowhere near enough to run complex algorithms.

Second, they are too fragile. The slightest environmental noise—a bit of heat, a vibration, even cosmic rays penetrating the Earth—can cause qubits to err. Once errors pile up, the computed results are useless.

Because of hardware limitations, scientists can only desperately fine-tune the "software (algorithmic circuits)," trying to complete computations with the fewest qubits and the shortest time.

But programming quantum circuits is completely different from our everyday code.

It has two insurmountable mountains:

1. No "If-then": Traditional programs can evaluate conditions and take route A or route B. But a quantum computer is a superposition of all possibilities; it must include both A and B operations in every run.

2. Absolutely no "garbage disposal": Traditional programs clear memory after using an intermediate result. But in the quantum world, information cannot be discarded, because quantum computation must be perfectly reversible! Forcefully erasing intermediate scratch work contaminates the result. Want to clear memory? The only way is to "play back" the computation in reverse, but that consumes expensive quantum gates.

This forces human scientists into an extreme tug-of-war between "space (qubits)" and "time (computation steps)."

Take simulating FeMoco as an example (this is a nitrogenase core, crucial for fertilizer production).

In 2017, humans estimated it would require 50 trillion Toffoli gates to compute. After 8 years of top scientists racking their brains, that number was only reduced to 340 million.

Humanity took 8 years. Claude took only 53 hours.

In 53 Hours, Mythos Created a Miracle

Anthropic set up a STEM Fellows project.

They gave Mythos 5 a task—

I'll describe in plain language 8 fundamental quantum circuits from chemistry, optimization, physics simulation, and number theory. You design them yourself.

The entire process was fully autonomous and closed-loop.

1. Claude used Python to write quantum circuit designs.

2. An automatic evaluation program checked correctness and gave a "score" (number of qubits used × number of quantum gates).

3. Claude received feedback and, with no human intervention, reflected and iterated on its own.

In just 53 hours, Claude beat humanity's best publicly available designs on all 8 tasks!

It directly overturned human thinking patterns, finding entirely new structures and mathematical laws.

Let's look at its two most explosive god-tier operations.

Operation One: Autonomously Deriving New Mathematical Formulas to Break a Deadlock

In number theory and cryptography (e.g., elliptic curve computations), there is a time-consuming core step called "point addition," whose most expensive operation is modular inversion.

Traditional computers do this quickly with an algorithm called "divsteps": repeatedly swap, add/subtract, and halve two numbers until the result emerges. In this process, traditional computers generate a lot of "branch information (scratch work)" and discard it after use.

But quantum computers cannot discard scratch work!

To run the divsteps algorithm on a quantum circuit, all historical branch information must be stored. This requires an astronomical number of qubits—completely impractical. Therefore, human scientists believed this path was a dead end.

But Claude didn't buy it.

It thought: If storing scratch work is too expensive, can we not store it and instead "reverse-engineer" the scratch work from the result?

At first, Claude was like a diligent student, blindly guessing a few rules, but when faced with 256-bit large numbers, it finally broke down.

Then came the hair-raising move: Claude abandoned brute-force testing and directly began pure mathematical theoretical derivation. It delved deep into the iterative process of divsteps, and hard-proved a series of entirely new algebraic identities!

Based on the new mathematical formulas, Claude wrote a precise decoding rule, even completing formal verification with Lean!

Ultimately, Claude produced the first quantum divsteps reversible implementation in human history that requires no storage of branch history.

Compared to human designs, it slashed the number of quantum gates by 17x while only increasing qubits by 12%!

Operation Two: Overhauling Google's Algorithm Architecture, Cutting Computational Cost by 12x

The second astonishing breakthrough occurred in the "Decoded Quantum Interferometry (DQI)" algorithm. This is quantum computing's most powerful weapon for logistics scheduling and materials design.

Simply put, the most costly step in the DQI algorithm is using a Reed-Solomon decoder for error correction.

Previously, Google experts designed a scheme using the "extended Euclidean algorithm."

But due to quantum circuits' rigid rule that they cannot take different branches, Google had to stretch the computation to an absurd number of rounds, wasting massive computational power.

After receiving the task, Claude glanced at Google's scheme and quickly concluded: This algorithm doesn't match quantum circuits; it needs to be replaced.

It directly swapped the core engine to the "Berlekamp-Massey decoding method." Its every round has identical operations, perfectly matching quantum circuits' rigid nature.

However, changing the algorithm brought a new "memory crisis"! Each round of computation generates a pile of temporary data that cannot be deleted—it must be kept for later "playback." If all were stored, 665 qubits would be needed, an astronomical figure.

Claude pulled off another god-tier move: it first used the same "reverse-derivation method" to eliminate most storage needs, leaving only a few "outliers."

If following humanity's most conservative approach, much space would still need to be reserved.

But Claude actually wrote its own high-speed simulator and ran a full 403 million sample tests to precisely measure the probability of anomalies occurring!

The result: it found that in the vast majority of cases, only 1 value needs to be stored.

So Claude simply left only 19 qubits of storage space in the circuit.

With this three-combo of "swap core algorithm + extreme memory reuse + big-data-powered precise space reservation," Claude's decoder design reduced gate count by a full 12x compared to the previous architecture!

What originally took 1 hour to compute can now be done in minutes.

Netizens Erupt: Is My Physics PhD Thesis a Joke?

Once this report came out, the internet exploded.

Someone said, "The predicted timeline for the quantum field is about to be completely upended. Humanity could get a useful quantum machine many years earlier!"

Someone lamented: We used to have AI explain existing science, but now, AI is discovering things we ourselves don't understand.

The most despairing part: perhaps every physics PhD has discovered that the paper they slaved over has an AI plain-language version, and it's 10x better than theirs.

A Milestone That Changes Human History Has Arrived

Mythos 5's achievement is truly exhilarating.

FeMoco nitrogenase simulation: humanity spent 8 years; Mythos 5 took only 53 hours. This means software is saving hardware!

We thought that to build a quantum computer capable of developing new drugs and producing clean fertilizers, humanity would need over a decade to overcome the "millions of qubits" hurdle.

But now, there's no need to wait for hardware upgrades—AI directly compresses algorithms tenfold or a hundredfold, and even a "broken machine" can produce revolutionary results!

Even more imaginatively, since every link in a quantum computer can be described, verified, and scored with code, Anthropic points out an alluring possibility—

Future AI will no longer optimize a single circuit in isolation, but conduct "global joint optimization" from the lowest-level hardware physical structure to the highest-level algorithmic logic!

Humanity will directly build hardware according to AI's blueprints.

The singularity of the technological explosion is much closer than we imagined.

References:

https://www.anthropic.com/research/designing-efficient-quantum-circuits

This article is from the WeChat public account "Xin Zhi Yuan" (ID: AI_era), author: ASI Apocalypse

Related Questions

QWhat did Mythos 5 reportedly achieve in the article's opening claim?

AMythos 5 reportedly autonomously closed the loop and designed new quantum circuits that were far more efficient than the human 'best solution,' based only on plain-language descriptions.

QWhy is designing quantum circuits especially difficult according to the article?

ABecause quantum circuits cannot use ordinary if-then branching and cannot discard intermediate information, since quantum computation must be reversible; this forces extreme trade-offs between space (qubits) and time (gate count).

QWhat example illustrates the dramatic speedup from human research to Mythos 5?

AFor simulating FeMoco, human estimates took 8 years to reduce the required Toffoli gates from 50 trillion to 340 million, while Claude reportedly accomplished its result in only 53 hours.

QHow did Mythos 5 improve the quantum divsteps approach for modular inversion?

AIt derived new algebraic identities, created a decoding rule without storing branch history, formalized it in Lean, and produced the first reversible quantum divsteps implementation that cut gate count by 17 times with only a 12% increase in qubits.

QWhat major optimization did Mythos 5 make to Google's DQI decoder architecture?

AIt replaced the extended Euclidean algorithm with Berlekamp-Massey decoding, used reverse-inference and extensive simulation to reduce memory needs, and cut the decoder's gate count by 12 times, from about an hour of computation to a few minutes.

Related Reads

Aave's Half-Year Report Card: A 97-Person Team Burned $20 Million in Half a Year, What Is the DeFi Leader Up To?

Aave Labs' 180-day report (Q2–Q3 2026) reveals a 97-person team that logged ~100,000 hours, completed 6,096 tracked items, shipped 28 product or protocol deployments, and spent nearly $19.48 million (excluding token compensation), while the Aave protocol generated about $44.58 million in revenue. Total deposits grew roughly 40% in Q3 to $33.8 billion, but this followed a sharp Q2 decline driven by the rsETH bridge exploit and a broader DeFi contraction, so the rebound was largely a partial recovery. Protocol revenue actually fell about 37% quarter-over-quarter, from $27.29 million to $17.29 million, as high Q2 liquidation fees (including Chainlink SVR) dropped sharply and borrowing interest income declined, even though GHO interest, swap fees, and stablecoin utilization improved. The report also signals Aave's expansion beyond crypto-native lending. Aave V4, launched in March, reached $1.34 billion in deposits by October 1, and now supports tokenized equities via an Equities Hub on Base, custody-based institutional lending with Anchorage and Chainlink, and externally operated independent instances such as EtherFi and Sentora that share revenue with the DAO. Aave Labs is also pursuing consumer and enterprise distribution through the Aave App (over $45 million in user balances), Stable Vaults for embedded yield, and an MCP Server letting AI assistants read protocol data and prepare—but not sign—transactions. Under the Aave Will Win Framework, the DAO funds development and receives gross brand-product revenue. The core open questions are growth quality, whether new businesses generate durable revenue, and balancing concentrated execution against DAO dependence on a key service provider.

marsbit57m ago

Aave's Half-Year Report Card: A 97-Person Team Burned $20 Million in Half a Year, What Is the DeFi Leader Up To?

marsbit57m ago

Circle Partners With DBS to Expand Institutional USDC Access Across Asia

Circle, the issuer of the USDC stablecoin, has partnered with Singapore-based DBS Bank to expand institutional access to USDC across Asia. Announced on November 18, the agreement allows eligible institutional clients to mint and redeem the dollar-backed stablecoin directly through DBS, with near-instant settlement between US dollars and USDC available 24 hours a day. This gives institutions another way to access USDC without relying entirely on cryptocurrency exchanges and other secondary-market liquidity providers, supporting corporate treasury operations, payment settlement and liquidity management. Institutions can convert between dollars and USDC without being restricted by conventional banking hours or payment cut-off times. Access is limited to eligible institutional clients rather than retail customers. The agreement builds on DBS and Circle's participation in Project BLOOM, an industry initiative led by the Monetary Authority of Singapore focused on developing infrastructure for the safe and scalable use of tokenised money. DBS reported that 88% of large corporations surveyed were already deploying or piloting tokenised money applications, with more than three in five adopting or planning to adopt stablecoins. Evy Theunis, head of digital assets at DBS's Institutional Banking Group, said the partnership would give institutions more choice in managing and transferring value, while Circle Chief Commercial Officer Kash Razzaghi said it would help expand institutional access to USDC across Asia. Transaction limits, fees and a detailed country-by-country rollout schedule were not disclosed.

TheNewsCrypto1h ago

Circle Partners With DBS to Expand Institutional USDC Access Across Asia

TheNewsCrypto1h ago

From Crypto VC to Traditional VC: The Three Essential Capabilities of Venture Capital

In this article, Big Brain Holdings investor Derrick Cui reflects on his transition from crypto venture capital to traditional VC in AI, reindustrialization, and space, contrasting crypto's short liquidity cycles and narrative-driven token prices with traditional VC's decade-long capital lockups and illiquid equity. He explains why LPs still allocate to venture despite poor average returns and illiquidity: each year produces rare companies capable of returning an entire fund; business success defies checklists, rewarding subjective judgment; consensus is often wrong, so non-consensus bets offer asymmetric upside; and top investors can repeatedly generate excess returns across cycles. He describes the investor's job as deal sourcing, diligence, and pitching, noting AI tools have lowered sourcing barriers but顶级 deals still come from genuine relationships, and that likability is an underrated professional asset. He then details three core capabilities: (1) Data—separating verifiable facts from founder promises, using third-party validation to pierce polished narratives; (2) Context—building comparative, industry-wide knowledge to judge talent and technical bottlenecks, now accelerated by reasoning models that shift the key skill from reading papers to asking higher-quality questions; and (3) Storytelling—constructing a rigorous "Right to Win" narrative that explains why a company prevails across multiple possible futures, sometimes via systematic cluster bets (e.g., SMRs for AI energy demand) rather than single-company wagers. He also admits past mistakes: being "nerd-sniped" by technical brilliance that doesn't equal a great business, chasing non-consensus ideas for their own sake, and rejecting the "spray and pray" mentality. Finally, he warns that fundraising incentives—DPI substitutes, markups, co-investment halo, hot-round chasing—can distort investment decisions, and concludes that VC is a trainable skill built on data capture, better questions, and credible winning narratives.

marsbit1h ago

From Crypto VC to Traditional VC: The Three Essential Capabilities of Venture Capital

marsbit1h ago

Trading

Spot
活动图片