Sovereign AI
Open bot
Open-source. Self-hosted. Publicly benchmarked.

The crypto-specialist LLM you can actually use.

Sovereign AI is a 14B-parameter open model fine-tuned on crypto markets, protocols, and on-chain patterns. It scores 87% A-rate on our 45-question benchmark — and you can call it from Telegram or REST in under a minute.

Open Telegram botSee benchmarksAPI docs →

Free tier: 3 queries/day, no credit card. Pro: $20/mo or 0.1 SOL/mo.

Live benchmark
v6-sft · promoted
87%
A-rate on 45 crypto Q&A
Sovereign v2 (14B)
87
GPT-4o (closed)
68
Claude 3.5 Haiku
64
Llama 3 70B
59
Qwen3-14B base
47
Comparisons are illustrative of our eval; run it yourself with the reproducibility script.

Open model, open benchmark

Model card, eval set, reproducibility script — all public. No marketing-only numbers.

Read the research →

Built for crypto specifically

Trained on DeFi mechanics, MEV patterns, tokenomics, Solana ecosystem, L2 designs. Ask domain-specific questions.

Try the bot →

Educational, not advisory

Every paid response carries a non-removable disclaimer. We explain, we don't call trades.

Read our disclaimer →

What the model does

/ask

General crypto Q&A

Example
/ask What's the difference between LST and LRT?
Response

Liquid Staking Tokens (LSTs) like JitoSOL represent staked SOL earning validator rewards. Liquid Restaking Tokens (LRTs) like those in EigenLayer or Symbiotic wrap LSTs and restake them for additional AVS rewards, at the cost of slashing exposure across multiple protocols…

/analyze

Deep-dive on a token or protocol

Example
/analyze JUP
Response

Jupiter is Solana's largest DEX aggregator by volume, with a native token (JUP) launched Jan 2024 via a 700M retroactive airdrop. Governance covers treasury allocation and DAO decisions. Risks include validator concentration, MEV exposure on aggregated routes, and TVL concentration in LP pools…

/predict

Scenario analysis, not price calls

Example
/predict SOL 4h
Response

Scenario framework for SOL 4h horizon. Bull case: BTC holds ATH, Solana ETF inflows >$100M/day, REV remains high. Bear case: macro risk-off, validator outage, L1 competition. Key levels to watch: 200-EMA, weekly open, Sanctum yield spreads. We don't call entries or exits — we flag the variables that matter…

/sentiment

Sentiment scoring of text

Example
/sentiment "SOL is getting crushed by macro, but fundamentals intact"
Response

{"score": -0.32, "label": "mildly bearish", "confidence": 0.71, "drivers": ["macro", "crushed"], "balanced_by": ["fundamentals intact"]}

Start with 3 free queries a day.

No credit card, no signup. Open the bot, send /ask, get answers grounded in a publicly benchmarked model.

Open Telegram bot