Roboflow

Muse Spark 1.2 vs Qwen3.5-27B

Muse Spark 1.2 averages 73.8% across five Vision Evals tasks and Qwen3.5-27B 64.1% across four (#26 and #40 of 73 models), so Muse Spark 1.2 leads overall.

  • Muse Spark 1.2 wins all 4 tasks both ran, with its widest lead on reasoning (75.1% vs 58.1%).
  • Qwen3.5-27B is about half the cost: $0.0043 vs $0.0083 overall average cost per sample.
  • Muse Spark 1.2 is faster: 11.5s vs 80.4s per sample.

Compare Muse Spark 1.2 vs Qwen3.5-27B live

Run the same image across every model that supports a task and compare their outputs side-by-side.

Detect and compare bounding boxes across models on the same image.

Open Object Detection in the full playground
MetaMuse Spark 1.2
Run to compare this model.
QwenQwen3.5-27B
Run to compare this model.

Muse Spark 1.2 vs Qwen3.5-27B on Vision Evals

Score by taskMuse Spark 1.2Qwen3.5-27B
Reasoning
75.1
58.1
Identification
89.6
80.2
Counting
76.6
67.6
Object Detection
59.0
50.5

Roboflow Vision Evals. Full breakdown below.

Muse Spark 1.2 vs Qwen3.5-27B Comparison Table

Evals updated November 19, 2026Pricing updated November 21, 2026

PropertyMuse Spark 1.2Qwen3.5-27B
OrganizationMetaQwen
Categoryclosedopen
Modalitymultimodalmultimodal
Release DateAug 2026Feb 2026
Context Window1.0M262K
ParametersUnknown27B
LicenseProprietaryApache 2.0
Commercial usecommercial use per vendor termsfree for commercial use
Pricing per 1M tokens
Input $/1M$1.25$0.195
Output $/1M$4.25$1.56
Vision Tasks
Both support 10 more tasks
Model Features
Both support 3 more features
Vision Evalsground-truth scores across 5 vision tasks, pooled at low effort
Overall
73.8%
64.1%*
Quantizationsself-hosted
BF1664.1%FP860.9%AWQ-INT461.2%hardware →
Avg cost / sample$0.0083$0.0043
Avg speed / sample11.51s80.37s
By task
Object Detection (low)
59.0%
±1.0, Mean of 3 runs, range 58.1 to 60.2
$0.0096
50.5%
±3.5, Mean of 3 runs, range 46.1 to 53.0
$0
Object Detection (high)
60.5%
±0.3, Mean of 3 runs, range 60.2 to 60.7
$0.014
–
Counting (low)
76.6%
±2.7, Mean of 3 runs, range 74.3 to 79.7
$0.0050
67.6%
±1.4, Mean of 3 runs, range 66.2 to 68.9
$0
Counting (high)
75.2%
±2.0, Mean of 3 runs, range 73.0 to 77.0
$0.0082
–
Identification (low)
89.6%
±1.6, Mean of 3 runs, range 87.5 to 90.6
$0.0038
80.2%
±4.7, Mean of 3 runs, range 75.0 to 84.4
$0
Identification (high)
87.5%
±0.0, Mean of 3 runs, range 87.5 to 87.5
$0.0062
–
OCR (low)
68.7%
$0.0086
–
by category
Single value
60.4%
Transcription
89.3%
Structured JSON
87.0%
Text localization
31.5%
OCR (high)
68.6%
$0.014
–
by category
Single value
61.7%
Transcription
85.1%
Structured JSON
86.7%
Text localization
30.7%
Reasoning (low)
75.1%
±0.3, Mean of 3 runs, range 74.8 to 75.5
$0.0073
58.1%
±2.3, Mean of 3 runs, range 55.6 to 60.3
$0
Reasoning (high)
75.7%
±0.3, Mean of 3 runs, range 75.5 to 76.2
$0.012
–

Muse Spark 1.2 vs Qwen3.5-27B: Overview

Muse Spark 1.2

Muse Spark 1.2 is a proprietary multimodal reasoning model from Meta Superintelligence Labs, released as a coding-focused update to Muse Spark 1.1. It accepts text, images, video, audio, and PDF documents and returns text, with a context window of roughly one million tokens that allows whole repositories, long documents, and extended agent trajectories to be held in a single request. The model thinks before answering, and the amount of reasoning effort it spends is configurable per request. Alongside its visual and document understanding, it supports structured output and parallel function calling, and it is designed to operate either as a planning agent that delegates work or as a subagent executing tasks in parallel.

Training for version 1.2 scaled up compute on coding tasks and widened the diversity of training environments, concentrating on long-horizon work such as whole-repository generation, large end-to-end projects, and automated research. Part of the training data was self-generated, with Muse Spark 1.1 producing coding environments and instruction-following templates and grading candidate solutions against them. The model was co-trained with the Muse Code terminal agent, incorporating rejection-sampled harness trajectories and that toolset. Multimodal use cases documented for the family include visual-to-code generation and detailed image and video captioning.

Qwen3.5-27B

Qwen3.5-27B is a multimodal dense hybrid model developed by Alibaba Cloud’s Qwen team and released in February 2026 as a high-precision entry in the Qwen3.5 "Medium" series. Unlike its Mixture-of-Experts (MoE) siblings, the 27B model utilizes a dense architecture combining Gated Delta Networks with a feed-forward structure, activating its full parameter suite for every inference to maximize reliability. The model features a native 262K-token context window, extensible to 1M tokens via YaRN (RoPE scaling), and is released under the Apache-2.0 license.

Optimized for agentic workflows, Qwen3.5-27B employs an early-fusion architecture that treats visual and textual data as a unified stream for deep cross-modal reasoning. While the larger MoE variants in the family lead in raw knowledge benchmarks, the 27B model offers a stable and high-density alternative for structured data extraction and spatial perception, contributing to the Qwen3.5 family’s generational leap in OCR accuracy over the previous Qwen3-VL series.