iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Yes—according to a TheoPhysEval arXiv paper submitted on October 9, 2026, GPT-6-Astra, running in Codex with literature-search and coding tools, produced accepted final answers for 49 of 50 unpublished theoretical-physics problems. That is 98% accuracy. The primary paper puts the average cost below $3 per problem; a contemporaneous Decoder report estimates approximately $2.40 per problem and just under $116 for all 50. The score is impressive evidence of reliable, checkable calculation, not proof that an AI independently chooses important questions or makes major physics discoveries.
What the 49-of-50 result means
The claim comes from TheoPhysEval: Assessing AI agents at autonomous theoretical physics problem solving, an arXiv paper submitted October 9, 2026. TheoPhysEval contains 50 unpublished problems with verifiable answers across 11 theoretical-physics subfields. Examples span fluid flows, biological processes, particle physics, cosmology and black-hole research.
| Measure | Reported result | Qualification |
|---|---|---|
| Final answers | 49 of 50 | Authors’ 2026 result for GPT-6-Astra harnessed in Codex with realistic tools |
| Accuracy | 98% | Equivalent to 49 accepted answers out of 50 |
| Average cost | Under USD 3 per problem | Primary paper’s figure |
| Estimated cost | About $2.40 per problem; just under $116 total | Approximate figures reported by The Decoder on October 15, 2026 |
| Coverage | 50 tasks in 11 subfields | Unpublished, checkable theoretical-physics calculations |
A 98% score means the agent’s submitted final results matched the benchmark’s accepted answers on 49 tasks. It does not mean every intermediate derivation was error-free, that the system can handle any physics problem, or that it solved 98% of all research in physics.
What TheoPhysEval actually tests
Concrete calculations based on known methods
The benchmark deliberately focuses on questions that apply established techniques to modified situations. The authors describe a typical problem as work a world expert could complete in about a day or less without AI, while a specialist from a neighboring field might need considerably longer. These are demanding research-grade calculations, but they are closer to advanced “turn-the-crank” work than to proposing a new theory.
#1 Best Overall
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
Why unpublished tasks matter
Contributors generally changed numerical values or conditions from calculations in their own papers. That makes copying a published answer harder while preserving a verifiable target. “Unpublished” therefore describes the benchmark questions, not a guarantee that no related method or equation appeared anywhere in the literature.
The realistic-tool setting
In the headline condition, Astra could search the literature and write scripts for numerical and symbolic work. That setup is intended to resemble how a capable researcher works with software and references, rather than testing a model in a text-only box.
Rank #2
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
Is the $2.40-per-problem figure real?
It is a secondary estimate, not the paper’s exact headline number. The authors state an average cost below $3 per problem. The Decoder’s October 15 report gives a more precise approximation of $2.40 per problem and just under $116 for the 50-problem set. Because $2.40 is rounded and the two sources report different levels of precision, it should be presented as an approximate operating cost, not a fixed price or a consumer rate card.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →The figure covers the benchmark run described in the report. It is not a quote for a guaranteed answer to a new physics question, and it does not include a human’s time to inspect assumptions, reproduce calculations or correct an answer.
Rank #3
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
How Astra compares with other agents
The Decoder also reported results for other agents on the same 50-task comparison. These numbers are secondary reporting and should not be treated as definitive until independently checked against the paper’s full evaluation details.
| Agent | Reported accuracy | How to read it |
|---|---|---|
| GPT-6-Astra | 98% (49/50) | Top result reported in the realistic-tool condition |
| Opus 5.5 | 92% | Secondary report; condition and grading details should be verified |
| Gemini 3.8 Flash | 68% | Secondary report |
| DeepSeek-V4.1-Flash | 54% | Secondary report |
| GLM 5.3 | 42% | Secondary report |
A fair comparison requires more than ranking percentages. Check whether every agent saw the same 50 tasks, had equivalent web and coding access, incurred comparable limits and costs, and was graded against the same independently checked references.
Rank #4
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Broad Compatible: The usb c to usb adapter widely supports all USB C/ USB A devices like laptops, tablets, cellphones, car chargers, and phone chargers. Such as compatible with MacBook Pro/Air 2023/2022, Thunderbolt 4/3 Devices,Apple MagSafe Watch 9/8/7/SE/Ultra, iPad Pro 2022/2021, Samsung Galaxy S23/S20/S10, and iPhone 17/16/15 Pro. Plug and play
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
Why 98% does not prove autonomous physics research
Solving a specified problem is different from selecting a problem
The benchmark supplies the question, the relevant physical context and a checkable target. Real research also requires deciding which question is worth pursuing, identifying a tractable path, recognizing when an assumption is physically wrong and judging whether a result is important. TheoPhysEval does not measure those choices.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsThe result is not evidence of a new theory or major discovery
The tasks exclude major open questions and emphasize known methods applied to altered inputs. A system can be excellent at deriving a requested quantity without inventing a useful framework, finding an unexplained phenomenon or deciding that an unexpected result deserves experimental follow-up.
Best Value
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
Tool ablation hints at memorized literature, but does not isolate the cause
The paper reports only small average accuracy changes when agents are denied web search and background papers. The authors interpret that pattern as evidence of substantial literature memorization. It is suggestive, not a clean separation of memorization, internal reasoning and tool use: removing a tool can change workflow, prompting and error checking at the same time.
How reliable is the 49-of-50 grading?
One task defeated every model
The report says one problem stumped all evaluated systems. That is a reminder that the 98% figure is an aggregate over a small, selected set; it does not establish a uniform success rate across subfields or difficulty levels.
Reference answers were revised after expert review
Expert follow-up found subtle errors in several reference answers. After those corrections, Astra’s responses were judged correct on the affected cases. The revisions improve the final assessment, but they also show why benchmark quality and independent checking are part of the result—not administrative details to ignore.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →What a robust replication should publish
- The exact 50 prompts, assumptions and permitted tools.
- Per-problem results rather than only the aggregate percentage.
- Reference solutions checked by independent domain experts.
- Separate scores for tool-enabled and tool-disabled conditions.
- Token, compute and human-review accounting behind the cost estimate.
What readers should conclude
The strongest defensible conclusion is narrow but significant: in a realistic software-and-literature workflow, GPT-6-Astra can produce correct answers to a broad sample of difficult, concrete theoretical-physics calculations at a reported average cost below $3 each. That makes it a potentially powerful assistant for derivations, numerical checks and literature-grounded problem solving.
The evidence stops short of showing autonomous science. TheoPhysEval does not test research taste, question selection, experimental design, theory invention or responsibility for deciding whether an answer is physically meaningful. Anyone using Astra for real work should still reproduce the mathematics, inspect units and assumptions, and have a domain expert review consequential conclusions.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
