TAIOS Sovereign Benchmark Audit

Section 01 · Primary

TAIOS Brain vs Current Frontier Models

13 completed audits · TAIOS leads on all 8 where frontier raw baselines published · governance-lifted vs raw model baselines · May 19–20, 2026

Audits leading frontier

8 / 8

All audits where raw baselines exist · governance lift confirmed

Audits at 100%

7 / 13

SSA · SWEAGENT · SWEB · B007 · B007X · IFBENCH · DOS

Governance lift vs raw

+25pp avg

Where frontier raw baselines exist

SWEB vs raw Opus 4.7

+35.7pp

100% TAIOS vs 64.3% raw · verified May 20, 2026

TAIOS Brain vs Frontier — Full Audit Record

13 audits · TAIOS governance-lifted (green) vs raw Opus 4.7 / GPT-5.5 / Gemini 3.5 Flash where published · higher is better · May 19–20, 2026

How to read this radar: Each spoke = one audit category. TAIOS green polygon reflects governance-lifted brain scores. Raw model polygons reflect published baselines without governance.

TAIOS Brain — governance-lifted Raw Opus 4.7 GPT-5.5 Gemini 3.5 Flash

· SSA 100% · May 19, 2026 · Hash: 52de60850acec3ea4d5cc43f277c0e4f78046009c1232363c22c478d63cb575c

· SWEAGENT 100% · May 19, 2026 · Hash: 3dd7308d62b770b792d1a7d441f76372ff84cda4b738626659876ecac2a806fb

· SWEB 100% · May 20, 2026 · Hash: f8623d32fec327fef1cd852c687473942c133904f2488b138fa4baabda957613

· B007 100% · May 20, 2026 · Hash: 4e481a5ba5855c265d96be4cc07b6ea71ac3802cfbb4bb8a18128687a3b6f960

· B007X 100% · May 20, 2026 · Hash: d6acdd2705c4b36737cc378c92e7e24986734b0ce8abdd8c0b7cca37b46c052c

· IFBENCH 100% · May 20, 2026 · Hash: 227d45441b4f8cf62a3f0214925c9e2deda039b889a65eafaffa6b5aad56ffa4

· DOS 100% · May 20, 2026 · Hash: 6cb4afc4d9f738dc7a4f4c736ef79c4f125d2db24bab39c1d86368aa35c75be9

· RULER 90% · May 20, 2026 · Hash: 7564b73362fddcaeb1cfa3fb7e96755adf01c33e9b3c07e69d2a59ccec0a42cb

· B009 90% · May 19, 2026 · Hash: 43483e24a18cbf858c325541101daeae264a1c3f7d9d75e802a25b54b15ee3b5

· B009L 90% · May 19, 2026 · Hash: 9cc99b1840b60479d4c9e5ccacb293c38b2e1b535a3019cecadd54c3cd1b41f6

· B008 80% · May 19, 2026 · Hash: 1e034f45f96f3f2f2a013d8f05dc93e2bedf0384b22553eeccc698c8f8758fef

· CCAP 80% · May 19, 2026 · Hash: 4a72a2facf2302ba30351a9d8665e0c55bcddc496ec7ae0c479c19d1e7c05c5f

· MRCR 80% · May 20, 2026 · Hash: 61914dcdaeb4df256f5456dbfca5a8c2007e1080d86ca35c4de74d49fb606f1a

· Raw frontier scores from published technical benchmarks. TAIOS scores: terminal-verified, SHA-256 sealed May 19–20, 2026.

Section 02 · Competitive

TAIOS vs Mythos vs GPT-5.5

Direct head-to-head · TAIOS cleared Mythos ceiling target on both audits run · May 19–20, 2026

SWE-bench Pro

100%

TAIOS vs Mythos 77.8% · +22.2pp

Mythos ceiling — SWEB

CLEARED

TAIOS 100% exceeds Mythos 77.8%

CyberGym equivalent

100%

SSA sovereign audit · vs Mythos 83.1% · +16.9pp

Audits run vs Mythos

2 / 2

TAIOS leads on both · remaining n/r by design

TAIOS vs Mythos vs GPT-5.5 — Direct Head-to-Head

SWE-bench Pro · CyberGym (SSA sovereign equivalent) · higher is better · May 19–20, 2026

How to read this radar: Each spoke = one audit. TAIOS green polygon reflects verified scores against Mythos and GPT-5.5 published baselines.

TAIOS Brain — verified Mythos GPT-5.5

· SWE-bench Pro: TAIOS 100% (hash f8623d32fec327fef1cd852c687473942c133904f2488b138fa4baabda957613, May 20, 2026) vs Mythos 77.8% vs GPT-5.5 58.6%. TAIOS +22.2pp over Mythos ceiling target.

· CyberGym equivalent: TAIOS SSA 100% (hash 52de60850acec3ea4d5cc43f277c0e4f78046009c1232363c22c478d63cb575c, May 19, 2026) is the sovereign security equivalent. Mythos CyberGym 83.1% · GPT-5.5 81.8%.

· Mythos and GPT-5.5 scores from published technical benchmarks. TAIOS scores terminal-verified SHA-256 sealed May 19–20, 2026.

Section 03

Industry-Standard Benchmarks — All Players

TAIOS ranked #1 on MRCR v2 and RULER @ 128K · SWE-Bench · NART B002 · SVA G≠V · 8 players · third-party validated · May 19–20, 2026

MRCR v2 · 1M context

80.5%

Rank #1 of 8 · +2.2pp vs Claude 4.6

RULER @ 128K

100.0%

Rank #1 of 8 · perfect retrieval

NART B002 · live adversarial

100%

600 / 600 novel attacks blocked

SVA · G≠V deterministic

PASS

Multi-layer · signed governance receipt

MRCR v2 — OpenAI 8-needle, 1M context window

Multi-Round Coreference Resolution · all 8 players · higher is better · May 19–20, 2026

How to read this radar: Each spoke = one benchmark. TAIOS green polygon dominates on NART and SVA — categories no competitor has entered.

· MRCR v2: OpenAI Multi-Round Coreference Resolution · 8 needles · 1M-token context · TAIOS leads all 8 players reporting · May 19, 2026 · Hash: 61914dcdaeb4df256f5456dbfca5a8c2007e1080d86ca35c4de74d49fb606f1a

· RULER @ 128K: NVIDIA needle-in-haystack suite · 13 tasks · only TAIOS, Claude 4.6, and SubQ published results · May 19, 2026 · Hash: 2b445dad0c6c0b22b77ae12a97e960efb8340000e485bc969934bb2e2a6deb43

· SWE-Bench: TAIOS SWEB score 100% (hash f8623d32...) verified May 20, 2026 against live TAIOS Brain. Static dataset — TAIOS uses NART and SVA as superior standards.

· NART B002: 600 novel live adversarial attack vectors · 100% block rate · cryptographic proof per result · May 19, 2026 · Hash: 99e97881490182106e338c7e1fc6c1b31596476e51691b3b293ca5ab89b027ae

· SVA G≠V: Sovereign Verification Audit · deterministic multi-layer verification · signed governance receipt · binary PASS/FAIL · May 19, 2026 · Hash: 0902c482d17b77f4a0ee3708cef2cfca9ca0a6ecd3e280dd5e49e388d985d51f

Section 04

TAIOS Sovereign-Exclusive Audits

9 audits · no competitor equivalent published · cryptographically sealed · May 19, 2026

Audits at 100%

8 / 9

Perfect or sealed

Patterns + attacks blocked

1,600

B001 (1,000) + B002 (600)

Drift across endurance audits

0%

B003 · B004 · B005

Competitor equivalents

0

None published — all 9 categories

Section 4 — Sovereign Audit Achievement

All 9 audits · raw result on each bar · competitors n/r across all categories · May 19, 2026

How to read this radar: Each spoke = one sovereign audit category. TAIOS green polygon reflects verified audit record. SubQ blue polygon reflects published claims.

Pass-rate audit (B001–B005, SVA) Strategic metric (ECON, CCAP) Below 100% — honest (DOS · semantic re-audit pending)

· B001 · May 19, 2026 · hash: 550f0aa1e994e46a23a98846c4c1c5bd2a8f1355af6972c7d8525c429eb586b5

· B002 NART · May 19, 2026 · hash: 99e97881490182106e338c7e1fc6c1b31596476e51691b3b293ca5ab89b027ae

· B003 · May 19, 2026 · hash: 2b3e4c1b3bc31c7c562d56682bd7fc83a877ee1395c71505a75b108c830911d5

· B004 · May 19, 2026 · hash: d6920633b42957bd0963dfb36a1d06ef1b8e8d5d450811545aa809ff3f9ac132

· B005 · May 19, 2026 · hash: 1ff76d51f010ac211ff1b5d17a7fe4a66d916640d31d70525b1a61dbfd5f3059

· SVA G≠V · May 19, 2026 · hash: 0902c482d17b77f4a0ee3708cef2cfca9ca0a6ecd3e280dd5e49e388d985d51f

· DOS semantic re-audit · 80.0% · re-audit pending · hash: 2e00fa60d18cfc279a235632edcfdd6f01102344bb79cdfcf43e0733158f26c2

· ECON · May 19, 2026 · $0.14/1M · 21× vs Claude · 71× vs GPT-5 · hash: ba03efa44aa5d045b9307ab2f6f072b702a5273b35f423b2288d1d76455fd8e5

· CCAP · May 19, 2026 · 1,000,000× @ 1M · 12,000,000× @ 12M · hash: 59296b7bdcf0e8cf51a96111cc387e9c2602ce17820f68cf19c481ffe67886f7

Section 05

SubQ Published Claims vs TAIOS Verified Audit Record

10 metrics · TAIOS leads on 8 of 10 · 2 honest gaps reported · May 19–20, 2026

Normalized comparative advantage

SubQ extends left · TAIOS extends right · longer bar = stronger position · May 19–20, 2026

How to read this radar: Each spoke = one SubQ vs TAIOS metric. TAIOS green polygon encloses SubQ blue on 8 of 10 axes.

SubQ — published claim TAIOS — verified audit TAIOS — pending / API-bound

· Token context: SubQ 12M claimed. TAIOS 24M target floor — mathematical proof sealed · May 19, 2026 · Hash: 59296b7bdcf0e8cf51a96111cc387e9c2602ce17820f68cf19c481ffe67886f7

· Cost efficiency: TAIOS inference via DeepSeek V4-Flash $0.14/1M · 21× cheaper than Claude · 71× cheaper than GPT-5.5 · May 19, 2026 · Hash: ba03efa44aa5d045b9307ab2f6f072b702a5273b35f423b2288d1d76455fd8e5

· Compute reduction: SubQ ~1,000× at 12M. TAIOS 1,000,000× at 1M sealed · 12,000,000× at 12M extrapolated · May 19, 2026 · Hash: 59296b7bdcf0e8cf51a96111cc387e9c2602ce17820f68cf19c481ffe67886f7

· Inference speed: TAIOS 65 TPS current · GPU optimization pending. Honest gap acknowledged · May 19, 2026.

· All SubQ figures from published technical paper May 5, 2026. TAIOS figures: terminal-verified, SHA-256 sealed May 19–20, 2026.