Official public profile • Rabbit Industries Pvt Ltd • updated 2026-09-30

Agastya AGI v01

Bramha Medha Native INDIA

REFERENCE POSITION #7PUBLIC CANDIDATEEVIDENCE R1 PUBLISHEDTHE FIRST WORLD NATIVE ENVIRONMENT

Ranking clarification: #7 is Agastya's position in Rabbit Industries' public reference list, using the same #1–#7 display format as the other systems. It is not an independently verified global performance rank. External competitive rank remains pending identical frozen benchmark runs and independent replay.

Published specifications

Model scale & training disclosure

Total parameters1.265BExact: 1,264,715,776
Active parameters1.265BExact: 1,264,715,776 • dense • 100% active
Canonical context4,096 tokens4.1K
Pretraining tokens1.000BExact current snapshot: 1,000,255,070
Architecture classDense nativeCanonical Native Brain
InitializationRandom initializationNo third-party weight initialization in canonical lineage
FamilyBramha Medha Native INDIASovereign native model family
Benchmark qualificationIN PROGRESSSame-suite external rank not yet established
Sovereign lineage

Canonical Native Brain path

Agastya corpus → Agastya tokenizer → Agastya architecture → RANDOM INITIALIZATION → Agastya pretraining → Agastya post-training → Agastya Native Brain. Reference Vault models are separated for evaluation and research; they are not weight ancestors of the canonical Native Brain.

Competition status

Reference Position #7 • external qualification pending

1Complete BM-1B capability acceptance on the frozen eligible native checkpoint.
2Freeze Benchmark-RC1: model version, harness, prompts, scoring, tool permissions and environment.
3Run identical public benchmark suites used for the frontier comparison wherever methodology is compatible.
4Independently replay the accepted benchmark package and preserve signed evidence.
5Publish methodology, results, hashes and caveats without filling missing scores with estimates.
6Only then assign an evidence-backed external competitive position or submit to an external leaderboard.
Current benchmark evidence

No fabricated frontier score

Agastya does not yet publish GPQA Diamond, Terminal-Bench 4.0, DeepSWE v1.1 or Humanity's Last Exam-with-tools scores from an identical frozen frontier-comparison harness. Those fields remain explicitly PENDING SAME-SUITE RUN. This prevents internal development diagnostics from being presented as directly comparable public benchmark results.

Public integrity policy

Facts, company designation and pending claims

Published facts: model/family name, dense 1.264715776B total and active parameters, 4,096-token canonical context, current 1,000,255,070-token training snapshot, and random-init sovereign lineage policy. Company designation: “The First World Native Environment.” Pending independent verification: global-first designation and same-benchmark competitive performance rank.

Competition referenceNative Brain Evidence R1Agastya Master PlanMachine-readable profile