← All models

IBM Granite 4.0 Micro

IBM

enterprise_transparent

Capabilities

ToolsCodeStructured OutputMultilingual

Strengths

  • Highest transparency score ever (FMTI 95)
  • Apache 2.0 — fully open
  • Full training data documentation
  • Enterprise audit trails
  • Part of Granite family (vision, code, guardian variants available)
  • MATH500 competitive with 24B models despite 8B size
  • Strong RAG with LoRA adapters

Weaknesses

  • Not frontier quality
  • No vision in base (separate Granite Vision models available)
  • Smaller ecosystem

Pricing

Input / 1M tokens

$0.017

Output / 1M tokens

$0.112

Context window

131k

Transparency

Open weights

10.0 / 10

Open training data

10.0 / 10

Open methodology

10.0 / 10

Licence openness

10.0 / 10

Provider disclosure

10.0 / 10

FMTI company score

950.0 / 10

Composite:10.0 / 10

FMTI 95/100 — highest ever. Perfect 100 on 10/14 dimensions. Full training data documentation. Apache 2.0.

Sustainability

Inference energy

7.0 / 10

Training footprint

7.0 / 10

Provider infrastructure

5.5 / 10

Composite:6.5 / 10

Published training compute/energy data. Smaller models run on consumer hardware. FMTI 95 includes disclosure on compute infrastructure.

Task Fitness

question answering
70%
code
78%
mathematics
75%
business communication
58%
analysis
72%
extraction
80%
long-form generation
68%
research
50%
reasoning
72%
summarisation
72%
translation
70%
conversation
68%