Leaderboard / Shanghai AI Laboratory (InternScience)

Sage-A1.5

Name: Sage-A1.5
Author: Shanghai AI Laboratory (InternScience)

Shanghai AI Laboratory (InternScience) · Sage · released 2026-10-15 · data checked 2026-11-03

Sage-A1.5 is a 36B-parameter Mixture-of-Experts model (3B active per token) from Shanghai AI Laboratory (InternScience) (China), released 2026-10-15 under the Apache-2.0 license. It doesn't have enough reported benchmarks for an index score yet. It needs about 23.2 GB of memory at 4-bit with an 8K-token context.

Agentic research model built on Qwen3.6 35B-A3B. The card reports agent and science benchmarks (GAIA 95.1, BrowseComp 81.7, HLE with tools 55.1, SWE-bench Pro 61.2) but none of the six core benchmarks, so it has no index score yet.

Small LM Index

—

Needs 2 core benchmarks (has 0)

Parameters

36B

3B active · MoE

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

262K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best medium model on that benchmark. Sources: 3 reported by Shanghai AI Laboratory (InternScience). How we label sources.

IFBench Lab-reported Instruction following81.6

AA-LCR Lab-reported Long context81.0

Class best: Nanbeige4.3 33B 84.0

SciCode Lab-reported Coding44.3

Specifications

Lab: Shanghai AI Laboratory (InternScience) (CN · China)
Based on: Qwen3.6 35B-A3B (fine-tune or distillation)
Architecture: Mixture-of-Experts
Total parameters: 36B
Active parameters: 3B
Size class: Medium
Context window: 262K tokens
Input modalities: text, image
Reasoning: Reasoning (thinking)
License: Apache-2.0
Openness: Open weights
Memory to run: ~23.2 GB at 4-bit, 8K context (more sizes)

Weights on Hugging Face ↗

Compare with

Sage-A1.5 vs Seed-OSS 36B Instruct (36.2B, index 60.3)
Sage-A1.5 vs Hermes 4.3 36B (36.2B, index 51.4)
Sage-A1.5 vs MiniCPM-V 4.7 35B-A3B (35.2B, index 54.5)

Sources

https://huggingface.co/InternScience/Sage-A1.5

Something wrong here? Report it.

Run it locally

Uses InternScience/Sage-A1.5-Q4_K_M-GGUF (Q4_K_M). Swap the tag for another size: Q4_K_M.

Ollama

ollama run hf.co/InternScience/Sage-A1.5-Q4_K_M-GGUF:Q4_K_M

llama.cpp

llama-server -hf InternScience/Sage-A1.5-Q4_K_M-GGUF:Q4_K_M

Then open http://localhost:8080 for a chat window.

MLX (Mac)

pip install -U mlx-lm
mlx_lm.chat --model mlx-community/Sage-A1.5-OptiQ-4bit

Text only; for images use mlx-vlm.

vLLM (GPU server)

vllm serve InternScience/Sage-A1.5

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	256K context
16-bit F16	73 GB	74 GB	78 GB
8-bit Q8_0	39 GB	40 GB	45 GB
4-bit Q4_K_M	23 GB	24 GB	28 GB

Check it against your GPU, Mac or phone →