Models

Arche 1.0

One model, two speeds. Arche 1.0 runs Kimi K3 open weights inside the Belna agentic harness — optimized for privacy and safety.

Model cardArche 1.0powered by Kimi K3

Foundation: Kimi K3 (Moonshot AI), open weights
Architecture: 2.8T-parameter mixture-of-experts, native vision
Context: 1M tokens
Harness: Belna agentic — approvals, sealed vault, sandbox, visible trace
Smart mode: Deepest reasoning — research, builds, reviews
Fast mode: Snappy everyday answers at lower cost

Safety layer: allowlisted tools, approval gates, masked secrets (sec_••••), sandboxed browser/code/computer use. The agent never holds your keys.

Benchmarks

Published technical results, Arche 1.0 (Kimi K3, max reasoning effort) against current frontier models. Bold marks the best score per row.

Benchmark	Arche 1.0	GPT-5.6 Sol	Claude Fable 5	Claude Opus 4.8
GPQA Diamond	93.5	94.1	92.6	91.0
Terminal-Bench 2.1	88.3	88.8	88.0	84.6
BrowseComp	91.2	90.4	88.0	84.3
OSWorld-Verified	84.8	83.0	85.0	83.4
SWE-Marathon	42.0	39.0	35.0	40.0
FrontierSWE	81.2	71.3	86.6	66.7
DeepSWE	67.5	73.0	70.0	59.0
ProgramBench	77.8	77.6	76.8	71.9
MCP-Atlas	84.2	83.6	84.7	83.6
HLE (with tools)	56.0	58.0	63.0	57.9
τ³-Banking	33.4	33.0	26.8	27.6

Source: Moonshot AI Kimi K3 technical results; competitor scores as reported in the same release (each evaluated in its own vendor harness). Arche 1.0 serves these weights through the Belna privacy-first harness.

Pricing

Free includes $10 of API costs. Pro $30 → $20 credit + $50 gift card. Max $50 → $50 credit + $100 gift card. Details on the pricing page.