ZMIME
Comparison · 3 models · Updated Nov 18, 2026

PaddleOCR VL vs MinerU2.5 2509 1.2B vs Granite-4.0-H-Micro

No independent capability benchmark covers every model here yet. The score uses price, inputs & features and context window.

Price in rating: Included · Exclude price

  1. PaddlePaddle

    PaddleOCR VL

    Released Oct 16, 2025

    64/100
    • ECI—
    • Price$0.02 / $0.02
    • Context131K
  2. Opendatalab

    MinerU2.5 2509 1.2B

    Released Sep 17, 2025

    58/100
    • ECI—
    • Price$0.024 / $0.071
    • Context16K
  3. IBM

    Granite-4.0-H-Micro

    Released Oct 2, 2025

    64/100
    • ECI—
    • Price$0.017 / $0.112
    • Context131K
01 — Verdict

Capability evidence is incomplete

There is not enough shared capability evidence to name an overall winner. The scores (PaddleOCR VL 64/100, Granite-4.0-H-Micro 64/100, MinerU2.5 2509 1.2B 58/100) compare the available specifications and price. The score weighs price 50%, inputs & features 30%, context window 20%. None of the shared benchmarks cover every model here yet, so capability is left out of this verdict.

  • CapabilityNot enough dataNo independent benchmark covers every model here yet
  • Lowest pricePaddleOCR VLPaddleOCR VL $0.02 · MinerU2.5 2509 1.2B $0.036 · Granite-4.0-H-Micro $0.041 per 1M tokens (3:1 blend)
  • Longest contextPaddleOCR VL and Granite-4.0-H-MicroPaddleOCR VL 131,072 · Granite-4.0-H-Micro 131,072 · MinerU2.5 2509 1.2B 16,384 tokens
  • Widest inputsPaddleOCR VL and MinerU2.5 2509 1.2BPaddleOCR VL: Text, Images · MinerU2.5 2509 1.2B: Text, Images · Granite-4.0-H-Micro: Text
  • Self-hostingAll open weightsEvery model here can be downloaded and run on your own hardware
How the score is built
MeasureWeightPaddleOCR VLMinerU2.5 2509 1.2BGranite-4.0-H-Micro
Price50%100100100
Inputs & features30%252525
Context window20%32032
Specs & price only100%64/10058/10064/100

Left out because at least one model lacks the data: capability. The remaining weights were rescaled.

02 — Side by side

Every spec in one table

Highlighted cells lead their row. Dashes mean the data is not published.

PaddleOCR VL vs MinerU2.5 2509 1.2B vs Granite-4.0-H-Micro specifications side by side
SpecificationPaddleOCR VLPaddlePaddleMinerU2.5 2509 1.2BOpendatalabGranite-4.0-H-MicroIBM
Capability
Capabilities Index (ECI)———
ECI rank———
Price per million tokens
Input$0.02$0.024$0.017 (best)
Output$0.02 (best)$0.071$0.112
Cached input———
Blended (3:1)$0.02 (best)$0.036$0.041
Long-context rateSame rateSame rateSame rate
Price sourceMedian of 1 providersMedian of 1 providersMedian of 1 providers
Limits
Context window131,072 tokens (best)16,384 tokens131,072 tokens (best)
Max output16,384 tokens—131,072 tokens (best)
Inputs and features
TextYesYesYes
ImagesYesYesNo
PDFsNoNoNo
AudioNoNoNo
VideoNoNoNo
ReasoningNoNoNo
Tool callingNoNoYes
Structured outputNoNoNo
Availability
WeightsOpenApache-2.0OpenAGPL-3.0Open
API model ID———
API providers111
ReleasedOct 16, 2025Sep 17, 2025Oct 2, 2025
Knowledge cutoff———
03 — Cost

What would a month cost?

Enter your expected volume in millions of tokens. List prices only; caching and batch discounts would lower these.

  • PaddleOCR VL$0.24
  • MinerU2.5 2509 1.2B$0.381
  • Granite-4.0-H-Micro$0.394
04 — Questions

Which should you choose?

Which is better: PaddleOCR VL, MinerU2.5 2509 1.2B or Granite-4.0-H-Micro?

There is not enough shared capability evidence to name an overall winner. The scores (PaddleOCR VL 64/100, Granite-4.0-H-Micro 64/100, MinerU2.5 2509 1.2B 58/100) compare the available specifications and price. The score weighs price 50%, inputs & features 30%, context window 20%. None of the shared benchmarks cover every model here yet, so capability is left out of this verdict.

Which is cheaper, PaddleOCR VL, MinerU2.5 2509 1.2B or Granite-4.0-H-Micro?

PaddleOCR VL is cheaper at $0.02 input / $0.02 output per million tokens (median across 1 API provider). MinerU2.5 2509 1.2B costs $0.024 input / $0.071 output per million tokens (median across 1 API provider); Granite-4.0-H-Micro costs $0.017 input / $0.112 output per million tokens (median across 1 API provider). At a typical mix of three input tokens to one output token, that is $0.02 per million tokens for PaddleOCR VL versus $0.036 for MinerU2.5 2509 1.2B (1.8× as much) and $0.041 for Granite-4.0-H-Micro (2× as much).

Which scores higher on benchmarks?

There is no independent benchmark that covers all three models yet. PaddleOCR VL has not been scored yet, MinerU2.5 2509 1.2B has not been scored yet and Granite-4.0-H-Micro has not been scored yet.

Which is better for coding?

There are no published SWE-bench Verified results for PaddleOCR VL, MinerU2.5 2509 1.2B and Granite-4.0-H-Micro yet, so there is no like-for-like coding score. Test both on a sample of your own tasks. Note that PaddleOCR VL and MinerU2.5 2509 1.2B does not support tool calling, which most coding agents need.

Which has the bigger context window?

PaddleOCR VL and Granite-4.0-H-Micro have the largest context windows (131,072 and 131,072 tokens), against 16,384 for MinerU2.5 2509 1.2B. Maximum output per response: PaddleOCR VL up to 16,384, Granite-4.0-H-Micro up to 131,072 tokens.

Which can read images, PDFs, audio or video?

PaddleOCR VL accepts text and images; MinerU2.5 2509 1.2B accepts text and images; Granite-4.0-H-Micro accepts text. PaddleOCR VL handles the widest range of inputs.

Are any of these open source?

Yes, all three publish their weights (Apache-2.0 and AGPL-3.0), so you can self-host them.

Which is newer?

PaddleOCR VL is the newest, released Oct 16, 2025. Granite-4.0-H-Micro came out Oct 2, 2025; MinerU2.5 2509 1.2B came out Sep 17, 2025.

How do you decide the winner?

This rating uses price (50%, blended price per million tokens (3 input : 1 output), log scale); inputs & features (30%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (20%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Nov 18, 2026.