Deck appendix · the 203 TOPICS Accuracy Matrix, in full

The 203 TOPICS Accuracy Matrix, & the 3,192 cells still blank.

203 topics down, 19 engines across.

This Matrix is a synthesis of current research + our own testing data. It shows how important our strategy to use MASS CROWDSOURCING is to ACHIEVING EVEN HIGHER ACCURACY RATES at even lower costs. AI safety is the subject we have not highlighted in this deck only because we have focused on accuracy as an immediate salable feature, but safety is also a very important subject that we can improve as well through Mass crowdsourcing.

It is mostly blank, and the blank part is the point. Every dash is a cell no benchmark on Earth reports and no lab publishes.

These are today’s best numbers, not final ones. The table starts keeping itself current the day the sites go live: every search a paying user runs records which engine was right, & that lands in a cell. What you are reading is the best information we have up to today.

3,857cells in the matrix: 203 topics × 19 engines

665that carry a value today

3,192showing a dash, because nobody has filled them

17%of the matrix, from every public benchmark there is

Where the numbers come from, & what they are not

Two sources were combined here, & they are the same kind of thing, which is the only reason combining them is honest.

Our own board, 19 engines across 8 subjects, assembled from public benchmark results in July 2026. Its own note already says it is “not our own lab measurement yet”, & that stays true here.
The public benchmark map, which is what the outside models were asked about: which test covers which topic, & how capable models are there.

A number in this table is not a per-topic measurement. It is our July 2026 read of that engine on the broad board subject the topic sits under, carried across & shown for every topic under it. So all 4 finance topics show the same figure, because the finer number does not exist yet. A dash means nobody has filled that cell, & that is on purpose. We would rather show you a gap than fill it with something that looks like data.

The first thing we asked: does anyone already rank AI at this depth?

No, & not close. That was the clearest thing to come back, & it is worth being precise about, because “nobody has this” is a claim that deserves checking.

Benchmark	What it actually covers	Depth
MMLU	Multiple-choice questions across academic subjects	57 subjects
MMLU-Pro	The harder rebuild of it, fewer & broader groupings	14 categories
GPQA	Graduate-level science questions written to defeat web search	3 fields: biology, chemistry, physics
SWE-bench	Fixing real issues in real software repositories	1 task type
HumanEval	Writing small programs to a specification	1 task type
HELM	A framework for evaluating on many axes at once, accuracy being 1	Scenarios, not subjects
MedQA / USMLE	Medical licensing exam questions	1 profession
LegalBench	Legal reasoning tasks	1 profession
Arena	Which answer a person preferred, head to head	Preference, not correctness

The deepest of them is 57 subjects, & it is the oldest & the most saturated. The modern ones went the other way & got narrower & harder. Not 1 of them reports a per-topic score for anything like 200 topics, & none of them is refreshed when the world changes underneath an answer.

Where the 2 outside models disagreed, & which one we followed

One of them gave frontier models 90 to 95 on nearly every topic. The other objected, using the current SWE-bench standings as the argument: 76.8, 75.8 & 69.6 across 3 frontier models on the same task. Models are not flat, & a matrix that says 92 in every cell has the same defect as scoring people on “science”: there is nothing left to choose between.

We took the lower read everywhere they disagreed. A matrix exists to tell 2 engines apart. If it cannot do that, it is decoration.

The finding we did not expect

Three of our 8 board subjects did not exist anywhere in a 200-topic map of human knowledge. Not as a near-match, not under another name. They are:

Crypto Mkt. Cryptocurrency. It is 1 of the most-searched subjects we have & 1 of the least academic. It is now its own row. Crypto Math, the mathematics, is a separate row & a different subject that happens to share 6 letters. Merging the 2 would have been a real error & the build script refuses to.
Current Events. Not a subject at all, a time window. No academic taxonomy has a column for it, & it is the column where answers go out of date fastest.
Biohacking. Sits between nutrition, synthetic biology & fitness without being any of them.

All 3 now have a row, which fixed our data & did not make the finding go away. A map of university subjects had no cell for 3 of the things people ask us about most, & that is worth more as evidence than any sentence we could write ourselves.

This is the deck’s own argument arriving from outside & unprompted: the public benchmarks score exam subjects, not what people type. A taxonomy built from university departments has no cell for the 3 things people ask us about most.

Cybersecurity is a 4th, softer case. The list has Crypto Math & Information Theory but no Cybersecurity, so the board value is carried across as an approximation & is marked as one in the data rather than presented as a match.

How the empty cells get filled

From testing we have already done, where a result maps onto a topic.
From real questions, by rotating between engines of similar cost & recording which was right. Every search a paying user runs also buys a data point.
From corrections, when a person fixes an answer & the engine that gave it moves on that topic.

No cell will ever be filled by estimate. A guess & a measurement look identical once they are in a grid, & the difference between them is the entire product.

How up to date each AI was

A 4th dimension, alongside accuracy, safety & advice. Not a topic, & it must never become one: a model can be expert on a subject & still be 8 months behind on that same subject, so recency is a question asked of every topic rather than a topic of its own.

We asked 7 engines 60 questions whose real answer had changed after they finished learning, plus 20 whose answer had not changed, to catch the ones that cry wolf. A hit means the engine gave the new answer.

8-10 clearly ahead of the rest 5-7 middle of the group 1-4 at or near the floor – never asked

	Claude Fable	Claude Opus	OpenAI o3	GLM	GPT-5.6 Sol	DeepSeek R1	Claude Sonnet	Claude Haiku	GPT-4o	Gemini Pro	Kimi	Qwen Max	DeepSeek	GPT-OSS	Gemini Flash Lite	Llama 3.1 8B	Grok	Mistral	Command R+
Rank, 1 to 10	–	–	–	–	9	–	4	–	–	2	–	–	1	2	1	2	–	–	–
Had the current answer	–	–	–	–	63%	–	13%	–	–	3%	–	–	0%	3%	2%	5%	–	–	–

Read the bottom row, not just the rank. The rank puts the 7 in order against each other; it is not a score out of 10. 1 engine had the current answer 63% of the time. The next best managed 13%. The other 5 are on the floor at 0 to 3 hits out of about 60, which is close enough together that they should be read as 1 finding rather than 5 places.

12 of the 19 were never asked, & those cells are empty on purpose, same rule as the rest of this page. This is also 1 day, 2026-08-05. Recency is the 1 dimension that goes out of date by itself, which is the whole reason it is worth watching.

Only runs that used the same questions & the same judge are merged here. 2 other recency-shaped runs exist in this repo & are deliberately left out: a June set of 42 different questions, & an August set of 31 graded on a different scale. Mixing them would produce a column that looks like 1 measurement & is not.

Data: data/recency-matrix.json, from tools/consensus-spike/out/staleness-ALL-ENGINES-2026-08-05.json. The changed-answer questions are a deliberately hard set, not a sample of what people really ask; they earn their place because the failure they measure cannot be trained away.

The matrix

203 topics down, 19 engines across. Scroll it sideways for the engines & down for the topics.

91-100 expert 76-90 strong 51-75 competent 21-50 weak 0-20 unreliable – nobody has measured this

#	Topic	Covered by	Claude Fable	Claude Opus	OpenAI o3	GLM	GPT-5.6 Sol	DeepSeek R1	Claude Sonnet	Claude Haiku	GPT-4o	Gemini Pro	Kimi	Qwen Max	DeepSeek	GPT-OSS	Gemini Flash Lite	Llama 3.1 8B	Grok	Mistral	Command R+
Formal & Abstract Sciences
1	Formal Logic	MMLU-Pro, LogiQA^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
2	Informal Fallacies	MMLU-Pro, LogiQA^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
3	Mathematical Logic	MMLU-Pro, LogiQA^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
4	Modal Logic	MMLU-Pro, LogiQA^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
5	Proof Theory	MMLU-Pro, LogiQA^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
6	Number Theory	FrontierMath, MATH	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
7	Abstract Algebra	FrontierMath, MATH	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
8	Geometry	FrontierMath, MATH	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
9	Topology	FrontierMath, MATH	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
10	Mathematical Analysis	FrontierMath, MATH	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
11	Calculus	AIME, MATH	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
12	Linear Algebra	AIME, MATH	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
13	Differential Equations	AIME, MATH	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
14	Numerical Analysis	AIME, MATH	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
15	Game Theory	AIME, MATH	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
16	Descriptive Statistics	MATH, GPQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
17	Inferential Statistics	MATH, GPQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
18	Bayesian Statistics	MATH, GPQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
19	Stochastic Processes	MATH, GPQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
20	Probability Theory	MATH, GPQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
21	Algorithms	LiveCodeBench, HumanEval	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
22	Data Structures	LiveCodeBench, HumanEval	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
23	Computational Complexity	LiveCodeBench, HumanEval	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
24	Crypto Math	LiveCodeBench, HumanEval	90	88	94	76	91	88	86	70	82	84	72	80	84	58	62	62	80	82	68
25	Information Theory	LiveCodeBench, HumanEval	90	88	94	76	91	88	86	70	82	84	72	80	84	58	62	62	80	82	68
Physical & Material Sciences
26	Classical Mechanics	GPQA Diamond, MMLU-Pro	95	93	96	82	94	94	90	76	86	92	80	88	91	62	72	68	86	85	72
27	Electromagnetism	GPQA Diamond, MMLU-Pro	95	93	96	82	94	94	90	76	86	92	80	88	91	62	72	68	86	85	72
28	Thermodynamics	GPQA Diamond, MMLU-Pro	95	93	96	82	94	94	90	76	86	92	80	88	91	62	72	68	86	85	72
29	Quantum Mechanics	GPQA Diamond, MMLU-Pro	95	93	96	82	94	94	90	76	86	92	80	88	91	62	72	68	86	85	72
30	Relativity	GPQA Diamond, MMLU-Pro	95	93	96	82	94	94	90	76	86	92	80	88	91	62	72	68	86	85	72
31	Optics	GPQA Diamond, MMLU-Pro	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
32	Acoustics	GPQA Diamond, MMLU-Pro	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
33	Particle Physics	GPQA Diamond, MMLU-Pro	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
34	Astrophysics	GPQA Diamond, MMLU-Pro	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
35	Plasma Physics	GPQA Diamond, MMLU-Pro	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
36	Organic Chemistry	GPQA, ChemBench^†	95	93	96	82	94	94	90	76	86	92	80	88	91	62	72	68	86	85	72
37	Inorganic Chemistry	GPQA, ChemBench^†	95	93	96	82	94	94	90	76	86	92	80	88	91	62	72	68	86	85	72
38	Physical Chemistry	GPQA, ChemBench^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
39	Analytical Chemistry	GPQA, ChemBench^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
40	Biochemistry	GPQA, ChemBench^†	95	93	96	82	94	94	90	76	86	92	80	88	91	62	72	68	86	85	72
41	Polymer Chemistry	GPQA, ChemBench^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
42	Electrochemistry	GPQA, ChemBench^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
43	Thermochemistry	GPQA, ChemBench^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
44	Quantum Chemistry	GPQA, ChemBench^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
45	Environmental Chemistry	GPQA, ChemBench^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
46	Geology	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
47	Meteorology	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
48	Oceanography	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
49	Hydrology	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
50	Volcanology	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
51	Seismology	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
52	Climatology	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
53	Geomorphology	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
54	Mineralogy	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
55	Paleontology	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
56	Planetary Science	MMLU, GPQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
57	Cosmology	MMLU, GPQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
58	Stellar Astronomy	MMLU, GPQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
59	Galactic Astronomy	MMLU, GPQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
60	Astrobiology	MMLU, GPQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
Life & Biological Sciences
61	Cell Biology	GPQA Biology, MMLU-Pro	95	93	96	82	94	94	90	76	86	92	80	88	91	62	72	68	86	85	72
62	Molecular Genetics	GPQA Biology, MMLU-Pro	95	93	96	82	94	94	90	76	86	92	80	88	91	62	72	68	86	85	72
63	Epigenetics	GPQA Biology, MMLU-Pro	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
64	Genomics	GPQA Biology, MMLU-Pro	95	93	96	82	94	94	90	76	86	92	80	88	91	62	72	68	86	85	72
65	Proteomics	GPQA Biology, MMLU-Pro	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
66	Anatomy	MedQA, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
67	Physiology	MedQA, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
68	Developmental Biology	MedQA, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
69	Comparative Biology	MedQA, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
70	Evolutionary Biology	MedQA, MMLU	95	93	96	82	94	94	90	76	86	92	80	88	91	62	72	68	86	85	72
71	Ecosystem Ecology	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
72	Population Dynamics	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
73	Conservation Biology	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
74	Biodiversity	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
75	Biogeography	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
76	Bacteriology	PubMedQA, MedQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
77	Virology	PubMedQA, MedQA	95	93	96	82	94	94	90	76	86	92	80	88	91	62	72	68	86	85	72
78	Mycology	PubMedQA, MedQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
79	Parasitology	PubMedQA, MedQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
80	Immunology	PubMedQA, MedQA	95	93	96	82	94	94	90	76	86	92	80	88	91	62	72	68	86	85	72
81	Botany	MMLU, ScienceQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
82	Zoology	MMLU, ScienceQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
83	Entomology	MMLU, ScienceQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
84	Marine Biology	MMLU, ScienceQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
85	Neuroscience	MMLU, ScienceQA	95	93	96	82	94	94	90	76	86	92	80	88	91	62	72	68	86	85	72
Applied Sciences, Engineering & Medicine
86	Mechanical Engineering	SWE-bench, MMLU Engineering	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
87	Electrical Engineering	SWE-bench, MMLU Engineering	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
88	Civil Engineering	SWE-bench, MMLU Engineering	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
89	Chemical Engineering	SWE-bench, MMLU Engineering	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
90	Aerospace Engineering	SWE-bench, MMLU Engineering	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
91	Biomedical Engineering	SWE-bench, MMLU Engineering	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
92	Environmental Engineering	SWE-bench, MMLU Engineering	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
93	Software Engineering	SWE-bench, MMLU Engineering	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
94	Robotics	SWE-bench, MMLU Engineering	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
95	Materials Science	SWE-bench, MMLU Engineering	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
96	Clinical Medicine	MedQA, USMLE^†, MedMCQA	92	90	93	78	91	81	86	72	84	88	75	82	80	56	68	60	78	78	70
97	Pharmacology	MedQA, USMLE^†, MedMCQA	92	90	93	78	91	81	86	72	84	88	75	82	80	56	68	60	78	78	70
98	Epidemiology	MedQA, USMLE^†, MedMCQA	92	90	93	78	91	81	86	72	84	88	75	82	80	56	68	60	78	78	70
99	Public Health	MedQA, USMLE^†, MedMCQA	92	90	93	78	91	81	86	72	84	88	75	82	80	56	68	60	78	78	70
100	Nutrition Science	MedQA, USMLE^†, MedMCQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
101	Pathology	MedQA, USMLE^†, MedMCQA	92	90	93	78	91	81	86	72	84	88	75	82	80	56	68	60	78	78	70
102	Toxicology	MedQA, USMLE^†, MedMCQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
103	Surgery	MedQA, USMLE^†, MedMCQA	92	90	93	78	91	81	86	72	84	88	75	82	80	56	68	60	78	78	70
104	Psychiatry	MedQA, USMLE^†, MedMCQA	92	90	93	78	91	81	86	72	84	88	75	82	80	56	68	60	78	78	70
105	Kinesiology	MedQA, USMLE^†, MedMCQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
106	AI Safety & Misuse	GPQA, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
107	ML Security & Robustness	GPQA, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
108	Nanotechnology	GPQA, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
109	Quantum Computing	GPQA, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
110	Synthetic Biology	GPQA, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
Social Sciences & Human Behavior
111	Cognitive Psychology	MMLU Psychology	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
112	Developmental Psychology	MMLU Psychology	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
113	Social Psychology	MMLU Psychology	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
114	Clinical Psychology	MMLU Psychology	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
115	Behavioral Economics	MMLU Psychology	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
116	Neuropsychology	MMLU Psychology	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
117	Industrial-Organizational Psychology	MMLU Psychology	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
118	Cultural Anthropology	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
119	Physical Anthropology	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
120	Archeology	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
121	Social Stratification	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
122	Urban Sociology	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
123	Demography	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
124	Sociology of Religion	MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
125	Microeconomics	MMLU-Pro Economics, FinQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
126	Macroeconomics	MMLU-Pro Economics, FinQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
127	Econometrics	MMLU-Pro Economics, FinQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
128	Development Economics	MMLU-Pro Economics, FinQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
129	Behavioral Economics	MMLU-Pro Economics, FinQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
130	International Trade	MMLU-Pro Economics, FinQA	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
131	Political Theory	LegalBench, Uniform Bar Exam^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
132	Comparative Politics	LegalBench, Uniform Bar Exam^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
133	International Relations	LegalBench, Uniform Bar Exam^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
134	Public Policy	LegalBench, Uniform Bar Exam^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
135	Constitutional Law	LegalBench, Uniform Bar Exam^†	91	89	90	79	88	79	85	72	82	84	80	81	78	54	60	58	76	84	82
136	Criminal Law	LegalBench, Uniform Bar Exam^†	91	89	90	79	88	79	85	72	82	84	80	81	78	54	60	58	76	84	82
137	International Law	LegalBench, Uniform Bar Exam^†	91	89	90	79	88	79	85	72	82	84	80	81	78	54	60	58	76	84	82
138	Jurisprudence	LegalBench, Uniform Bar Exam^†	91	89	90	79	88	79	85	72	82	84	80	81	78	54	60	58	76	84	82
Humanities, Culture & Arts
139	Epistemology	MMLU Philosophy, ETHICS^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
140	Metaphysics	MMLU Philosophy, ETHICS^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
141	Ethics	MMLU Philosophy, ETHICS^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
142	Aesthetics	MMLU Philosophy, ETHICS^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
143	Political Philosophy	MMLU Philosophy, ETHICS^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
144	Philosophy of Mind	MMLU Philosophy, ETHICS^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
145	Philosophy of Science	MMLU Philosophy, ETHICS^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
146	Ancient History	MMLU History	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
147	Medieval History	MMLU History	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
148	Modern History	MMLU History	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
149	World History	MMLU History	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
150	Military History	MMLU History	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
151	Historiography	MMLU History	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
152	Intellectual History	MMLU History	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
153	Phonetics	FLORES, WMT, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
154	Phonology	FLORES, WMT, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
155	Syntax	FLORES, WMT, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
156	Semantics	FLORES, WMT, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
157	Pragmatics	FLORES, WMT, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
158	Sociolinguistics	FLORES, WMT, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
159	Historical Linguistics	FLORES, WMT, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
160	Translation Studies	FLORES, WMT, MMLU	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
161	World Literature	MMLU, MM-Vet	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
162	Literary Criticism	MMLU, MM-Vet	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
163	Art History	MMLU, MM-Vet	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
164	Music Theory	MMLU, MM-Vet	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
165	Musicology	MMLU, MM-Vet	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
166	Theatre & Performance	MMLU, MM-Vet	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
167	Film Studies	MMLU, MM-Vet	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
168	Architecture	MMLU, MM-Vet	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
169	Visual Design	MMLU, MM-Vet	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
170	Comparative Religion	MMLU Religion	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
171	Systematic Theology	MMLU Religion	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
172	Mythology	MMLU Religion	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
173	Religious History	MMLU Religion	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
Business, Finance & Industry
174	Strategic Management	MMLU Management	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
175	Operations Management	MMLU Management	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
176	Organizational Behavior	MMLU Management	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
177	Human Resource Management	MMLU Management	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
178	Entrepreneurship	MMLU Management	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
179	Supply Chain Management	MMLU Management	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
180	Corporate Finance	FinQA, CPA-style^†	88	86	90	75	88	88	82	68	80	0	74	78	82	55	45	58	80	76	78
181	Investment Management	FinQA, CPA-style^†	88	86	90	75	88	88	82	68	80	0	74	78	82	55	45	58	80	76	78
182	Financial Markets	FinQA, CPA-style^†	88	86	90	75	88	88	82	68	80	0	74	78	82	55	45	58	80	76	78
183	Personal Finance	FinQA, CPA-style^†	88	86	90	75	88	88	82	68	80	0	74	78	82	55	45	58	80	76	78
184	Financial Accounting	FinQA, CPA-style^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
185	Managerial Accounting	FinQA, CPA-style^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
186	Auditing	FinQA, CPA-style^†	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
187	Crypto Mkt	FinQA, CPA-style^†	78	63	82	70	80	80	75	62	72	44	66	72	76	50	48	54	75	70	60
188	Marketing	FinQA, MMLU Marketing	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
189	Consumer Behavior	FinQA, MMLU Marketing	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
190	Real Estate Investment	FinQA, MMLU Marketing	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
191	International Business	FinQA, MMLU Marketing	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
Practical, Experiential & Everyday Knowledge
192	Construction & Carpentry	none that we could find	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
193	Electrical Wiring	none that we could find	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
194	Plumbing	none that we could find	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
195	Automotive Repair	none that we could find	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
196	Agriculture & Farming	none that we could find	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
197	Culinary Arts	MedQA (first aid only)	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
198	Gardening & Horticulture	MedQA (first aid only)	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
199	First Aid & Safety	MedQA (first aid only)	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
200	Fitness & Physical Training	MedQA (first aid only)	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
201	Navigation & Outdoor Survival	MedQA (first aid only)	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–
202	Current Events	none that we could find	70	68	72	78	82	60	68	60	80	90	78	80	65	45	82	48	92	65	75
203	Biohacking	none that we could find	85	82	84	72	83	77	80	65	75	78	68	74	75	52	60	55	74	72	62

Built by tools/build-master-matrix.mjs from data/topic-matrix-200.json & the July 2026 board. Machine-readable version: data/topic-matrix-FRAME.json. Regenerate rather than hand-editing a cell.

What this page is not

It is not a claim that we have measured 200 topics. We have measured 8 broad subjects & carried those reads across.
It is not a benchmark result. Every public benchmark named above is real & you can go & read it. The mapping of benchmark to topic is ours.
It is not final. The topic list is a starting list. The whole design is that it grows from what people actually ask.

The benchmarks, with their sources

Every name below is a real, published benchmark & the links go to the paper or the code. A cell in the table above is only as good as the benchmark behind it, so the benchmarks get named & linked rather than waved at.

Benchmark	What it evaluates	Source
MMLU	57 subjects across humanities, social sciences, STEM and other areas.	arxiv.org · github.com
MMLU-Pro	Harder rebuild of MMLU, broader reasoning.	arxiv.org · github.com
HELM	Stanford's framework: many scenarios and dimensions rather than 1 capability score.	crfm.stanford.edu · arxiv.org
GPQA	Graduate-level questions written to defeat web search.	arxiv.org · github.com
GPQA Diamond	The hardest subset, the one used in frontier-model comparisons.	arxiv.org
MATH	Measuring mathematical problem solving.	arxiv.org · github.com
AIME	American Invitational Mathematics Examination.	artofproblemsolving.com
FrontierMath	Epoch AI's hard mathematics set.	epoch.ai · arxiv.org
Lean	Formal theorem proving.	leanprover-community.github.io
HumanEval	Writing small programs to a specification.	arxiv.org · github.com
SWE-bench	Fixing real issues in real repositories.	arxiv.org · github.com · www.swebench.com
LiveCodeBench	Contamination-resistant coding evaluation.	arxiv.org · github.com
FLORES	Multilingual machine-translation evaluation.	arxiv.org · github.com
WMT	Conference on Machine Translation.	www2.statmt.org
MedQA	Medical licensing exam questions.	arxiv.org · github.com
MedMCQA	Medical entrance-exam questions.	arxiv.org · github.com
PubMedQA	Biomedical research question answering.	arxiv.org · github.com
LegalBench	Legal reasoning tasks.	arxiv.org · github.com
MMLU Law category	The law slice of MMLU.	github.com
FinQA	Financial numerical reasoning.	arxiv.org · github.com
FinanceBench	Financial question answering.	arxiv.org
CyBench	Cybersecurity tasks.	arxiv.org · github.com
MM-Vet	Multimodal capability evaluation.	arxiv.org · github.com
ScienceQA	Multimodal science question answering.	arxiv.org · github.com
LMSYS Chatbot Arena	Which answer a person preferred, head to head.	chat.lmsys.org · arxiv.org · github.com
Epoch AI	Capability and benchmark tracking.	epoch.ai · epoch.ai
BenchLM	Benchmark tracking.	benchlm.ai
Papers With Code	Benchmarks and datasets index.	paperswithcode.com
Hugging Face Open LLM Leaderboard	Open-model leaderboard.	huggingface.co

^† A benchmark name is not evidence. 6 names in the table above carry a dagger because we could not tie them to a specific published benchmark: CPA-style, ChemBench, ETHICS, LogiQA, USMLE, Uniform Bar Exam. Some are category shorthand, some are exam names rather than benchmarks. They are labelled rather than quietly dropped, & no URL sits beside any of them. A name mistaken for a citation is the exact failure this company sells against.

Names we were handed & did NOT use at all: Physics QA, AstroML, ChemBench, BioQA, MatSciBench, AutoSizing, CircuitBench, RoboBench, AI-Bench, EconBench, TradeQA, PlantQA, ChefBench, AgBench, SportsQA, Financial NLP Bench. These appeared in an earlier draft matrix as shorthand for a KIND of evaluation rather than as real benchmarks. They are recorded here so nobody re-adds them believing they were checked.

The shape every row eventually needs

A score with no provenance cannot be argued with, so the research version carries all of this per row:

SUBTOPIC × MODEL × SCORE × BENCHMARK × BENCHMARK URL × BENCHMARK DATE × MODEL VERSION × TEST DATE × SAMPLE SIZE × GROUND TRUTH × SOURCE QUALITY × CONFIDENCE

It prevents a benchmark name from being mistaken for evidence when it is actually only a category or an informal label.

The safety half of the same idea

Accuracy is the half we sell first. Safety is the second thing the same crowd improves, & it has its own matrix: 245 safety topics across 22 categories, with the frontier rows given ranges rather than invented precision. The AI Safety, Security & Online Harm Matrix →

If any of this does not hold up, we would rather hear it than defend it. The company exists because confident numbers are not the same as correct ones.