Free tools Windows power users keep installed
One-click scans. No signup required.
In Debashish Ghosal’s v0.1.0 AdversarialDebate field test, DeepSeek + Mistral led the reported average score and verdict rate—but Ghosal also reported a 65% capitulation rate for that pair. The apparent winner therefore raises a measurement problem: a system can record a debate as resolved whether participants reached agreement through argument or one side conceded almost immediately. The figures below are Ghosal’s reported results, not an independently replicated benchmark.
Why the highest score did not mean the most reliable pair
AdversarialDebate is a multi-agent debate and review system. Its reported outcomes combine results such as a score or convergence measure, a verdict rate, and concessions. Those outcomes can look strong without showing how the agents got there. As Ghosal puts it, “Strong-looking multi-agent metrics are not automatically trustworthy multi-agent metrics.”
As an Amazon Associate I earn from qualifying purchases.
In the v0.1.0 field test, DeepSeek + Mistral had the highest reported average score, 0.982, and the highest verdict rate, 97%. Yet Ghosal reported that 65% of debates involving this pair were capitulation cascades. A high rate of recorded verdicts could therefore coexist with a process in which one side yielded rather than substantively testing the other’s position.
Ghosal’s rule classified a debate as a capitulation cascade when at least 80% of concessions occurred in round one and there were zero rebuttals. Across 411 debates in v0.1.0, 80 were classified this way, or 19%. GPT + Gemini had a reported 0% cascade rate, but that pair’s verdict rate was only 4%—a reminder that avoiding capitulation is not the same as reaching useful resolution.
#1 Best Overall
- 50 millimeter focal length and maximum aperture of f/1.8
- Great for portraits, action, and nighttime photography; Angle of view (horizontal, vertical, diagonal): 40º, 27º,46º
- Minimum focusing distance of 1.15 feet (0.35 meter) and a maximum magnification of 0.21x
- Stepping motor (STM) delivers near silent, continuous move Servo AF for movies and smooth AF for stills
- 80 millimetre effective focal length on APS C cameras, 50 millimetre on full frame cameras. Lens construction: 6 elements in 5 groups
What the v0.1.0 pair results show—and leave out
Ghosal’s v0.1.0 results are reported in the source article, published August 29, 2026. The table keeps the reported aggregate score, verdict rate, and concession count together. These figures describe that version’s field test; they do not establish a universal ranking of model pairs.
| Pair | Average score | Verdict rate | Concessions |
|---|---|---|---|
| DeepSeek + Mistral | 0.982 | 97% | 2,352 |
| GPT + Mistral | 0.754 | 48% | 1,728 |
| GPT + GPT | 0.688 | 57% | 1,444 |
| Gemini + DeepSeek | 0.622 | 10% | 1,470 |
| Gemini + Mistral | 0.512 | 4% | 1,073 |
| GPT + Gemini | 0.357 | 4% | 727 |
The figures alone do not show how many concessions were supported by evidence, whether a rebuttal was substantive, or how results varied across tasks. The cascade rate adds one process signal, but it is a threshold-based classification, not a complete measure of debate quality. A pair can fail by refusing to converge; another can look successful by converging too easily.
How the recommendation changed across versions
Later versions changed both the comparison set and the interpretation. Their numbers should not be merged into a single leaderboard: v0.2.0 compares a full corpus with a validation subset, while v0.2.1 adds a comparison on the same 150-artifact corpus.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteRank #2
- EF Mount; Aperture Range: f/4-45; DC Autofocus Motor; 4.9' Minimum Focus Distance; 58mm Filter Thread Diameter
- 4.9-foot closest focusing distance; 32- to 8-degree diagonal angle of view
- Measures 2.8 inches in diameter and 4.8 inches long; weighs 16.8 ounces
- Improved mechanism makes zooming smoother; front part of zoom ring sports silver ring.
| Version and context | Pair | Reported result |
|---|---|---|
| v0.2.0, full corpus | GPT + Mistral | Average convergence 0.536; 2/150 verdicts; 2,927 concessions |
| v0.2.0, validation subset | DeepSeek + Mistral | Average convergence 0.572; 1/36 verdicts; 936 concessions |
| v0.2.0 | GPT + Gemini | Average convergence 0.033; 0/24 verdicts |
| v0.2.1, same 150-artifact corpus | GPT + Mistral | 0.536 |
| v0.2.1, same 150-artifact corpus | DeepSeek + Mistral | 0.572 |
| v0.2.1, same 150-artifact corpus | GPT + GPT | 0.273 |
| v0.2.1, same 150-artifact corpus | DeepSeek + GPT-4o-mini | 0.246 |
| v0.2.1, same 150-artifact corpus | GPT + Gemini | 0.033 |
In v0.2.0, Ghosal described GPT + Mistral as the full-corpus default and DeepSeek + Mistral as a validation pair. That is a change from the headline implication of v0.1.0, not confirmation that one combination is always best. The v0.2.0 figures also show why context matters: the DeepSeek + Mistral score came from a 36-item validation subset, not the 150-item full corpus.
In v0.2.1, Ghosal added DeepSeek + GPT-4o-mini on the 150-artifact corpus. The reported score of 0.246 was below the reported 0.536 for GPT + Mistral and 0.273 for GPT + GPT. Ghosal interpreted the comparison as evidence that Mistral’s participation mattered more than simply mixing labs. But the following update made that interpretation less certain.
Why the later gap is not decisive
In v0.2.2, Ghosal described the 0.572 versus 0.536 difference between DeepSeek + Mistral and GPT + Mistral as 1.8 sigma. That is a narrow separation, not strong grounds for a categorical recommendation. He also raised shared RLHF conversational defaults as a competing explanation: similarities in how models are trained or tuned could shape their interactions, rather than a uniquely positive effect from Mistral.
Rank #3
- Compact, lightweight and high-image quality RF tele zoom lens, with a versatile zoom range of 100-400mm
- Optical Image Stabilizer with up to 5.5 Stops of shake correction
- Up to 6 stops of shake correction when paired with EOS R series cameras featuring In-Body Image Stabilizer (IBIS)
- Minimum focusing distance of 2.89 feet at 200mm and maximum magnification of 0.41x at 400mm
- High speed, smooth and quiet autofocus with Canon’s Nano USM
Ghosal described results for pairs with fewer than 30 observations as having very wide noise floors. The v0.2.0 GPT + Gemini result, for example, was based on 24 cases. Its low score is a signal from that limited sample, not a precise estimate of how the pair would perform generally.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
The v0.2.1 release note reproduced in Ghosal’s article reports a 1.7–3.4% missed-issue rate, described there as first recall data, and 55 new unit tests. Those software-development details do not independently validate the pair comparison or settle which interaction pattern produces better review.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What a useful multi-agent comparison should measure
A single convergence or verdict score compresses different behaviors into one outcome. Ghosal frames the issue this way: “You cannot trust pair-level success metrics unless you also inspect how that success was produced.” A more informative evaluation reports outcome, process, sample context, and uncertainty side by side.
Rank #4
- High image quality and bright f/2.8 aperture zoom RF L lens
- Optical image stabilization of up to 5 Stops of shake correction
- High speed, smooth and quiet autofocus with Nano USM
- Min. Focusing distance of 0.69 ft by 0.21M (wide), 1.25 ft. by 0.38M (tele)
- A control ring for direct setting changes
- Convergence: how often, and how strongly, the agents move toward a shared conclusion.
- Verdict rate: how often the system records a final decision; a verdict alone does not establish that the reasoning was sound.
- Capitulations: how often one agent concedes, especially early in the exchange.
- Rebuttals: whether the interaction includes counterargument rather than immediate acceptance.
- Corpus and subset: the number and kind of artifacts behind each result, and whether a score is full-corpus or validation-only.
- Uncertainty and noise: the statistical separation between pair results and the instability expected from small samples.
These measures answer different questions. Low convergence can indicate persistent disagreement or a weak review protocol; high convergence paired with immediate concessions can indicate premature agreement. Neither pattern, by itself, proves that the system reliably catches errors. The v0.2.1 release note’s reported recall figure is relevant to missed issues, but it is a separate measure and does not resolve the pair-ranking uncertainty.
What this field test can—and cannot—establish
Ghosal’s account is a developer’s field-test report, not a controlled independent benchmark. The article does not establish exact prompts, provider settings, precise model snapshots, costs, or a fully reproducible protocol. Model labels such as “GPT” and “DeepSeek” should therefore be read as labels used in that project’s reported runs, not as timeless product specifications.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →The project’s version history is still useful as a case study in evaluation design: a headline metric can reward outcomes while hiding whether the system earned them. The v0.1.0 results make the issue vivid, while v0.2.2 cautions against turning a small score gap into a broad claim about model providers or pairings.
Ghosal’s article and its reported figures: The Best Model Pair in My Field Test Was Also the Least Trustworthy.
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.
