#31BFeb 28, 2026Evidence B

BAGEL-2

Name: BAGEL-2 multimodal review
Item: BAGEL-2
Rating: 76.9
Author: CN Multimodal Tier List

ByteDance · PKU · Undisclosed (unified MoT)

An open unified model balancing understanding and generation

BAGEL led comparable unified models on MMMU, MMBench and MM-Vet while supporting editing; documents and video are clearly weak.

TextImageNative

Last updated Oct 18, 2026Editorial scoring

Bottom line

For teams studying unified understanding-generation, or needing a hackable open editing base. Production stability and Chinese support still need validation.

Eight-dimension breakdown

editorial mapping · 0–100

BAGEL-276.9

Visual perception & grounding20%

Document parsing & Chinese OCR18%

Chart & document reasoning14%

Video & long-horizon reasoning14%

GUI agents & tool use14%

Generation & editing6%

Chinese context & culture6%

Ecosystem, cost & availability8%

Weights: perception 20 / documents 18 / charts 14 / video 14 / agents 14 / generation 6 / Chinese 6 / ecosystem 8. Definitions and sources on the methodology page.

Public benchmarks

as reported, not converted

Benchmark / metric	Value	Note
MMMU / MMBench	领先	Among unified models
模态	2	Image + text

Sources:

BAGEL 开源仓库 BAGEL 论文

Public scores come from different versions and prompts and are not interchangeable with our composite; they are one evidence input.

Scenario mapping

five stress tests, one scale

Scenario	Score	Observation
Chinese receipt OCR		76 · Weak
Chart data extraction		74 · Weak
30-min video QA		68 · No long video
Mobile GUI control		65 · Unsupported
Chinese cultural context		78 · Average

Strengths and limitations

Strengths

Unified understanding and generation with high research value.
Hackable open weights with decent editing.
Leads understanding benchmarks among unified models.

Limitations

Weak video and documents.
Under-optimized for Chinese scenes.
High deployment and inference cost.

Price and real cost

List price

—N/A

Open, self-hosted. For 1,000 images at 1,024px, ~1,100 image tokens each and 300 output tokens, the monthly cost is about ¥0.0 (assuming input at 25% of output price).

Use your volume

About BAGEL-2

Is BAGEL-2 open source?

Licence: Open (research license). Weights are available, but check the exact terms and filing requirements before commercial use.

How much does BAGEL-2 cost?

Open, self-hosted. It ranks #31 with a composite of 76.9. Real bills depend on resolution, tiling, caching and volume discounts.

What is BAGEL-2 best for?

Unified-model research; Open image-editing base; Academic reproduction。

Why does BAGEL-2 score 76.9?

The composite weights eight dimensions: perception 20%, documents 18%, charts 14%, video 14%, agents 14%, generation 6%, Chinese 6%, ecosystem 8%. Its strongest dimension is Generation & editing; its weakest is GUI agents & tool use.

Personal ranking · Not an official score

Every ranking, score and comparison on this page is an independent personal ranking, not an official score, and represents no vendor, institution or regulator. Scores are a subjective mapping of public evidence under our own weights — change the weights and the order changes. Disagree, or spotted a factual error? Contact the site owner: jizhi2024@foxmail.com — we check each report and log corrections in the changelog.