Models¶

The Macaron hosted API serves the Macaron-V1 model family. All variants are built on a Mixture-of-LoRA (MoL) architecture: a frozen base model plus a small set of specialist LoRA adapters, with a router selecting one specialist per user turn. See What is Macaron for the architecture and routing loop.

Hosted lineup¶

Model ID	Aggregate params	Base	Specialists	Use
`macaron-v1-venti`	748B	744B GLM-5.2 (frozen)	4 × 1B LoRA (L0 Chat, L1 Agent, L2 Coding, L3 GenUI)	Flagship; chat, agent, coding, GenUI via routing
`macaron-v1-tall`	50B	Qwen3.6-35B-A3B (frozen)	4 × 1B LoRA (same four-specialist design)	Local / lower-latency deployment
`macaron-v1-coding-venti`	748B	744B GLM-5.2	coding LoRA merged into base (single-specialist)	Coding-specialized; no routing overhead
`glm-5.2`	744B	GLM-5.2	none (base only)	Base model without MoL specialists

Routing¶

The two MoL-served variants (macaron-v1-venti, macaron-v1-tall) route each user turn through a three-stage loop: route (L0 classifies the request into one specialist under a 24-token budget), answer (the chosen specialist responds), summary (the specialist emits a ≤192-token summary stored server-side as shared context). macaron-v1-coding-venti and glm-5.2 skip routing: the coding LoRA is merged into the base, and glm-5.2 is the base alone.

Pricing¶

Effective per-token pricing, sourced from the dashboard /api/v1/models/pricing endpoint on 2026-08-06. Pricing is subject to change; verify current rates in the dashboard.

Models & Pricing — the dashboard Inference tab lists each model with final input, cache-input, cache-storage, and output per-million-token rates.

Model	Input ($/1M tok)	Output ($/1M tok)	Cache write ($/1M tok)	Cache read ($/1M tok)
`macaron-v1-venti`	7.00	25.00	2.00	2.00
`macaron-v1-tall`	1.80	10.00	0.00	0.20
`macaron-v1-coding-venti`	5.20	18.20	1.30	1.30
`glm-5.2`	5.20	18.20	1.30	1.30

macaron-v1-coding-venti and glm-5.2 carry a 0.65 effective rate multiplier against the base rate; macaron-v1-venti and macaron-v1-tall are at the base rate. Cache-write cost for macaron-v1-tall is zero.

Self-hosted base models¶

The lineup above is the hosted Macaron family. If you are self-hosting MinT and serving Qwen3 community base models on your own endpoint, see the self-hosted models reference for the Qwen3 lineup. The two surfaces are distinct: the hosted API serves Macaron models over /v1/chat/completions; the self-hosted surface serves Qwen3 base models over /oai/api/v1.