Agastya AGI v01 — Model Card R1

**Publisher:** Rabbit Industries Pvt Ltd

**Family:** Bramha Medha Native INDIA

**Release:** Public Evidence R1 — 2026-09-30

**Public evidence:** https://zippri.com/agastya-native-brain-evidence/

Model summary

Agastya AGI v01 is a dense native foundation-model program developed under the Rabbit Industries Native Forge sovereignty rule.

Canonical lineage:

Agastya corpus -> Agastya tokenizer -> Agastya architecture -> RANDOM INITIALIZATION -> Agastya pretraining -> Agastya post-training -> Agastya Native Brain

Published technical facts

• Total parameters: 1,264,715,776
• Active parameters: 1,264,715,776
• Architecture density: dense
• Canonical context: 4,096 tokens
• Tokenizer: ANTR2-R1, BPE, 98,320 vocabulary entries, NFC normalization, ByteLevel pre-tokenization
• Step-0 provenance: fresh random initialization
• Teacher weight ancestry: ZERO
• Third-party weight ancestry: false
• Third-party tokenizer artifact import: false
• Clean-P0 R2 final acceptance: PASS

Evidence boundaries

R1 publishes cryptographic identifiers, validation and acceptance manifests. It does not redistribute the proprietary raw checkpoint, full tokenizer vocabulary, raw corpus or architecture source code.

Evaluation status

Same-suite frontier benchmark qualification and independent external replay remain pending. No universal or independently verified global performance rank is claimed by this model card.

Intended use

Research, evaluation and development of the Agastya Native Brain program and compatible Rabbit Industries products, subject to production safety and human-gated deployment policy.

Limitations

The currently published evidence package is company-generated and externally citable, but not yet independently third-party verified. Public benchmark parity should not be inferred from internal development diagnostics.

Citation

See CITATION.cff in the Public Evidence R1 package.