Harvard, Anthropic, and QuEra Use LLMs to Discover Quantum Circuits with Up to 9,100x Lower Resource Costs

quantumcomputingreport.com Oct 17, 2026

Harvard, Anthropic, and QuEra Use LLMs to Discover Quantum Circuits with Up to 9,100x Lower Resource Costs

Curated by Shubham · TechCapsules · Oct 17, 2026

AI-summarised brief · reviewed before publication

Researchers from Harvard University, Anthropic, and QuEra Computing have unveiled an evolutionary AI framework that leverages large language models (LLMs) to automatically generate and optimize fault‑tolerant quantum circuits. The system, detailed in an arXiv paper, pairs Claude‑style LLM code proposals with a classical evaluator that tests candidates using the kickmix model and scores them by a combined Toffoli‑count‑and‑qubit‑footprint metric (T×Q‑cost). Compared with manually crafted baselines, the framework reduces T×Q‑cost by factors ranging from 4.1‑fold to over 9,100‑fold across several benchmark tasks, including a new reversible architecture for prime‑field modular inversion and a Hermitian code decoder that cuts peak qubits from 190,000 to 2,395. The approach discovers novel co‑design principles and replaces costly operations with algorithmic shortcuts, delivering orders‑of‑magnitude resource savings.

💡 Why It Matters

· By automating the discovery of ultra‑efficient quantum circuits, the method accelerates the path toward practical, large‑scale quantum computing, turning previously infeasible algorithms into near‑term candidates.

Read full article on quantumcomputingreport.com

Jev is Vercel’s fastest-adopted AI model ever

Vercel’s AI gateway data shows Jev, a lightweight decision‑making model from TypeSafe AI, reached 35% adoption among paid teams within two weeks of its September 15 release, nearly double the uptake of the next most popular model, Sonnet 5.5. Jev’s limited‑output design allows faster, cheaper processing, making it suitable for large‑scale data classification. Vercel, an early adopter, used Jev to replace GPT‑6 Luna for harmful‑command filtering. Open‑source models now process over 70% of gateway tokens, while frontier‑lab spending has increased tenfold.

💡 Jev’s rapid uptake demonstrates that streamlined, cost‑effective models can quickly dominate enterprise usage, reshaping how companies balance open‑source and proprietary AI solutions.

techcrunch.com · Oct 12

Meta releases its first frontier model, Muse Wonder

Meta’s Superintelligence Lab unveiled two new frontier models, Muse Wonder and Muse Wonder Full, marking the company’s first release of large‑scale AI systems. Unlike the earlier lightweight Muse Spark, the Wonder models are designed for higher capability, achieving benchmark scores comparable to Fable 5.1 and GPT‑6 Astra and outperforming rivals on the MemoryBench continual‑learning test. Meta emphasizes “active memory,” claiming the models can retain user information over extended periods. Access is provided via the Meta Model API at $4 per million input tokens and $20 per million output tokens, a price point similar to GPT‑6 Sol and Opus 5.5. Trained at the Prometheus data center in Ohio, the launch follows a 2023 restructuring of Meta’s AI division and follows the release of Muse Spark and Muse Agent.

💡 By delivering a model that combines high performance with long‑term memory, Meta aims to differentiate its AI offerings for personal assistants, challenging the dominance of OpenAI and Anthropic in the frontier‑model space.

appleinsider.com · Oct 9

You can run AI models on that Intel Mac Pro

The open‑source ToshLLM project enables developers to run large language models locally on Intel‑based Mac Pro machines equipped with AMD GPUs, leveraging Apple’s Metal graphics API for native acceleration. By bypassing cloud services, the framework eliminates per‑token usage fees and allows users to exploit existing hardware rather than purchasing Apple’s M‑series silicon. ToshLLM advertises “zero cloud, native Metal acceleration, and no per‑token bill,” positioning itself as a cost‑effective alternative to Apple’s M5 Ultra‑powered Mac Studio, which offers higher performance but requires Apple‑designed chips. The solution targets professionals and hobbyists who already own AMD graphics cards, promising a functional, though slower, AI inference pipeline that runs entirely on‑premises without additional licensing or subscription costs.

💡 It democratizes access to on‑device AI by turning existing Intel‑Mac hardware into a viable inference platform, lowering the barrier for developers who cannot afford or do not want to migrate to Apple’s proprietary silicon.

9to5mac.com · Oct 8

OpenAI brings GPT-6 to ChatGPT and debuts Intelligent UI

OpenAI rolled out two new GPT‑6 models for ChatGPT, expanding its generative AI suite. GPT‑6 Sol, tuned for everyday conversation, is now available to Plus, Pro, Business and Enterprise users, while GPT‑6 Luna will reach Free and Go subscribers starting tomorrow. Both models incorporate an “Intelligent UI” that dynamically blends text, images, charts, buttons and other interactive widgets according to the user’s query and the response format. The interface can present side‑by‑side comparisons, interactive diagrams, calculators, bill‑splitting tools and even simple games directly within the chat. Additionally, GPT‑6 can begin answering while still processing, adding information incrementally rather than waiting for a complete reasoning chain. Early tests report quicker replies and vivid visual aids. OpenAI detailed these features in a blog post.

💡 Embedding interactive tools turns the chat experience into a versatile workspace, letting users compute, visualize and experiment without switching apps.

business-standard.com · Oct 6

Nvidia-backed Reflection unveils open AI model to rival Chinese models

Reflection AI, a startup founded by former Google DeepMind researchers and backed by Nvidia, launched its open‑weight large language model named Beam on Monday. Beam is marketed toward enterprise users and developers for coding and AI‑agent applications, and the company says it delivers performance comparable to China’s flagship GLM‑5.2 on advanced reasoning benchmarks while costing less than many rival offerings. The firm announced billion‑dollar computing agreements with Elon Musk’s SpaceX and cloud provider Nebius Group NV to support model development. Reflection, which emerged from stealth last year, has been courting a $2.5 billion financing round at a $25 billion valuation. The release aligns with U.S. efforts to promote sovereign AI exports and counter the growing availability of inexpensive Chinese open models globally.

💡 Beam gives U.S. firms a cost‑effective, open alternative that could shift enterprise AI procurement away from Chinese‑origin models.

xda-developers.com · Oct 4

I set up Claude Code the way Spotify does to save tokens,…

Joe Rice‑Jones, a long‑time tech writer now at XDA, experimented with token‑saving techniques for AI‑driven coding tasks after learning about Spotify engineers’ “AiKA Modes” in their internal Portal platform. He recreated the two modes using Claude Code and several other large language models, aiming to reduce the token consumption that accrues when the API is billed per usage. While the recreated setup did lower token usage for simple prompts, the overall cost rose because the configuration required additional processing steps and higher‑tier model calls. Rice‑Jones details the trade‑offs he encountered, noting that the savings were offset by increased complexity and the need for more expensive model instances, ultimately making the approach less economical than anticipated.

💡 The experiment shows that mimicking proprietary AI optimizations can backfire, exposing hidden costs that outweigh the promised efficiency gains.