harn-vm 0.10.153

Async bytecode virtual machine for the Harn programming language
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27

# ── QC defaults ──────────────────────────────────────────────────────────────
# Default low-cost model per provider for cheap quality-check / repair
# passes. Scripts read these via `qc_default_model(provider)`.

[qc_defaults]
anthropic = "claude-haiku-4-5-20251001"
openai = "gpt-6-luna"
openrouter = "google/gemini-2.5-flash"
ollama = "llama3.2"
# `local` previously pointed at gpt-4o, a hosted OpenAI model that the
# bundled local server does not serve. Use the
# bundled vLLM Gemma 4 route instead.
local = "gemma-4-26b-a4b-it"
mistral = "mistral-small-2603"
cohere = "command-a-plus-05-2026"
xai = "grok-build-0.1"
# Groq retired qwen/qwen3.6-27b (absent from its live /models index on
# 2026-09-20). Qwen3.8 27B is Groq's only remaining Qwen route, at a higher
# rate ($0.80/$4.00 per MTok versus $0.60/$3.00). If the QC rate matters more
# than staying in the Qwen family, groq/openai/gpt-oss-20b is catalogued at
# $0.075/$0.30.
groq = "qwen/qwen3.8-27b"
minimax = "MiniMax-M2.5-highspeed"
# GLM-4.6 is Z.AI's cheap coding route, confirmed serving at paas/v4.
zai = "glm-4.6"
deepseek = "deepseek-flash"
gemini = "gemini-3.5-flash-lite"