Sealevel / v1.0 / solana language model

The Solana coding AI that actually knows Solana.

7B model fine-tuned on 270K Solana records from 500+ repos. 91% on Solana benchmarks, 100% HumanEval. Chat, explain transactions, decode errors — in your browser, terminal, or editor.

Try it now $ pip install slm-cli
At a glance
Solana eval
91%
192 / 210 tasks · 13 categories
HumanEval
100%
20/20 general coding
Base
Qwen2.5-Coder-7B
QLoRA · rank 32
Training
270K
from 500+ open-source repos
~ slm · chat
$ slm chat "derive a PDA for a vault"
→ thinking…

// anchor 0.30+ — constraint-based derivation
#[derive(Accounts)]
pub struct InitVault<'info> {
  #[account(
    init, payer = authority,
    seeds = [b"vault", authority.key().as_ref()],
    bump, space = 8 + Vault::INIT_SPACE,
  )]
  pub vault: Account<'info, Vault>,
Overall score
91%
192 / 210 tasks · 13 categories · Apr 2026
Base modelQwen2.5-Coder-7B (dense)
MethodQLoRA SFT · r=32
Data270K records · 500+ repos
HumanEval100% (20/20)
01Anchor Constraints15/15
100%
02CPI Patterns10/10
100%
03Error Handling10/10
100%
04SPL Token Ops10/10
100%
05Tx Construction9/10
90%
06PDA Derivation13/15
87%
07Adversarial6/10
60%

Pick your surface.

$ pip install slm-cli
$ slm config --api-key slm_xxx
$ slm chat "how do i derive a PDA?"

→ Streaming from slm-solana (7B)…
# use Anchor 0.30+ constraint-based derivation
#[account(seeds = [b"vault"], bump)]

Python CLI

Drop into any terminal. Streaming, syntax highlighting, session history. Works with any OpenAI-compatible endpoint.

  • pypi: slm-cli
  • Commands: chat · gen · review · migrate
  • History at ~/.slm/history
  • MIT license