Model reference · open weights

Macaron-Tall

Available as managed deployment LLMs mindlab-research Text gen 1 variants 555 dl/mo

Macaron-Tall is an open-weight language model from mindlab-research. AxForge deploys and operates it for you on dedicated EU-owned hardware — with the licence handled where one is required.

Available as managed deployment — configured and operated for you on dedicated EU hardware, quoted per deployment.

What it is

Released bymindlab-research
TypeLanguage models
TaskText gen
Parameters (lead)36.0B
Context256k tokens
Runs withtransformers
Released2026-07-22
Popularity555 downloads / month
LicenceOpen weights

About

What Macaron-Tall is

📖 Blog: Introducing Macaron-V1 🚀 Hosted API: Macaron API Platform 🧩 Artifacts: Macaron Artifacts 🛠️ Self-hosted serving: Mixture of LoRA (MoL) serving harness 📄 Technical report: Macaron-V1: Towards Open Continual Learning with Self-Improvement and Mixture-of-LoRA (arXiv:2608.09819) ✉️ Correspondence: contact@mindlab.ltd

Macaron-V1-Tall is a Mixture of LoRA (MoL) model in the Macaron-V1 family, built on Qwen3.6-35B-A3B for personal intelligence, tool use, coding workflows, and code-native Generative UI.

The model combines a Qwen3.6 MoE base with the Macaron-V1 specialist system: four LoRA specialists for chat, personal-agent tasks, coding, and GenUI, with an L0 router selecting the most suitable specialist for each new user request.

Macaron-V1-Tall is released alongside Macaron-V1-Venti as a sibling model in the Macaron-V1 family. Both models share the same product-facing specialist layout and routed serving interface while targeting different deployment and experimentation profiles.

Read the full model card

Highlights

  • A Qwen3.6-35B-A3B-based Macaron-V1 model with the full Macaron specialist system.
  • Mixture of LoRA (MoL) architecture with four specialists across Chat, Agent, Coding, and GenUI.
  • Built for personal-agent workflows, tool use, repository-level coding, and UI4A Generative UI.
  • BF16 checkpoint with four LoRA adapters stored under loras/L0 through loras/L3.
  • Supports a 262K context length according to the released checkpoint configuration.

Model Overview

FieldValue
Model nameMacaron-V1-Tall
OrganizationMindLab Research
Release familyMacaron-V1
Base modelQwen3.6-35B-A3B
ArchitectureQwen3.6 MoE base + Mixture of LoRA (MoL) specialists
Parameter footprint50B release label: a nominal 35B Qwen3.6-35B-A3B base plus four LoRA specialists (~50.1B by nominal-base-plus-adapter count)
LoRA rank / alphar = 64, α = 128, dropout 0
Stored values per adapter3,775,651,840 (BF16 for L0/L1/L3; F32 for L2)
SpecialistsL0 Chat, L1 Agent, L2 Coding, L3 GenUI
Post-training systemMinT + MindForge
Primary domainsPersonal intelligence, tool use, coding, Generative UI
Context length262K
Text config40 layers, 2048 hidden size, 16 attention heads, 2 KV heads, 256 experts, 8 experts per token
Precision / serving formatBF16 base checkpoint with routed LoRA serving
LicenseMIT

On parameter counts. Tall's adapter configurations use a different rank and expert target set from Macaron-V1-Venti, so its aggregate is reported as a rounded release footprint rather than inferred from base size alone. The two adapter budgets are not directly comparable because the bases expose different target-module and expert structures. The sidebar figure on this page is an automatic safetensors tensor count and may differ from this architectural figure.

Mixture of LoRA (MoL) Architecture

AdapterRoleDescription
loras/L0ChatConversational and instruction-following backbone; entry point for routing.
loras/L1AgentPersonal-life agent tasks, heavy tool use, long-horizon planning, and dynamic workflows.
loras/L2CodingCode understanding, SWE tasks, terminal use, and repository workflows.
loras/L3GenUIUI4A rendering and UI-driven action.

At runtime, L0 routes each new user request to the most suitable specialist. Ongoing reasoning and tool interactions remain within the selected LoRA, while completed work can be shared across specialists through concise summaries.

Routing Behavior and Cost

HopAvg latencyShare of loop
Route (L0 constrained decode, 24 tok)0.20 s11%
Answer (specialist generation)1.24 s70%
Summary (192-tok cap)0.32 s19%
Total1.76 s100%

Measured over 48 multi-turn mixed-domain requests at temperature 0. The routing overhead share (~30%) is stable across base sizes, while absolute loop latency is roughly 2.7× lower than Macaron-V1-Venti's 4.68 s.

On the same 6,448-sample trace used for Venti — verified identical by matching every sample ID and input hash — Tall reaches 99.04% routing accuracy with 100% canonical-label compliance, so routing quality is stable across base sizes. L3 GenUI reaches 100% and L1 Agent is lowest at 95.1%. The trace is drawn from LoRA training data, so this is an implementation diagnostic rather than a generalization estimate.

Evaluation

Macaron-V1-Tall compared with its Qwen3.6-35B-A3B base on the seven benchmarks evaluated for both systems. Both use the same protocol within each row; higher is better, normalized to 0–100.

BenchmarkMacaron-V1-TallQwen3.6-35B-A3B
Macaron ChatBench54.948.0
Macaron LivingBench48.447.1
PinchBench86.282.5
ClawGym64.058.6
SWE-bench Verified75.473.4
TerminalBench 2.156.252.5
UI4A-Bench (Final Score)59.333.9

Tall has a larger point estimate than its base on all seven rows, from +1.3 on Macaron LivingBench to +25.4 on UI4A-Bench. This is an end-to-end system comparison, not a parameter-matched component ablation: the two systems differ in parameter footprint, adapters, routing, and potentially harness behavior. The table does not identify which component produces each difference, and it does not establish scaling behavior from Macaron-V1-Tall to Macaron-V1-Venti.

The headline Macaron-V1 comparison against frontier baselines is reported for Macaron-V1-Venti; see the Venti model card and Table 8 of the technical report. Per-benchmark judges, case counts, and metrics are documented in Appendix B of the report.

Multimodal Behavior

The Macaron-V1-Tall specialists are trained on text-only data, while the Qwen3.6-35B-A3B base is multimodal. The following full-dataset point estimates (no-thinking mode) co

From the published model card. Full card on the HuggingFace links in the sidebar.

Benchmarks

Reported results

As published on the model card — the maker's own numbers, not measured by AxForge.

TaskDatasetMetricScore
ChatChatBenchChatBench54.900
ChatLivingBenchLivingBench48.400
AgentPinchBenchPinchBench86.200
AgentClawGymClawGym64
CodingSWE-bench VerifiedSWE-bench Verified75.400
CodingTerminalBench 2.1TerminalBench 2.156.200
Generative UIUI4A-BenchUI4A-Bench59.300

Using it via the API

Call it like any OpenAI endpoint

Once AxForge deploys macaron-tall for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (macaron-tall below is illustrative; you get the exact model name on deployment.)

$ curl -sS https://api.axforge.ai/v1/chat/completions \
  -H "Authorization: Bearer $AXFORGE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"macaron-tall","messages":[{"role":"user","content":"Hello"}]}'

Create an account — your API key is available in the console. 3M free tokens every 30 days with every new account.

© 2026 AxForge · EU-hosted AI infrastructure Pricing Docs Trust Privacy Terms