Skip to content
← Models
Open weightsIBM/10 trustVerified

Granite 4.0 H Small

Hybrid Mamba-2/Transformer MoE language model, open under Apache 2.0.

Visit official page →

About

Granite 4.0 H Small is IBM's flagship open-weight language model of the Granite 4.0 generation, released October 2025. It uses a hybrid architecture that mixes Mamba-2 state-space layers with a minority of transformer attention layers in roughly a 9:1 ratio, with 32B total parameters and 9B active per token. IBM reports the hybrid design cuts memory requirements for long-context and multi-session inference by over 70 percent versus conventional transformers.

The model ships under the Apache 2.0 license with cryptographically signed checkpoints, and IBM describes Granite as the first open model family to receive ISO 42001 certification for AI management. It targets enterprise workloads where throughput and memory efficiency matter more than raw scale, with native support on vLLM and other common runtimes.

Specs

Context window

Pricing

Free quota

API access

Weights

Open weights

Released

October 2, 2025

Benchmarks

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

enterprise-chatlong-context-reasoningragcode-assistance

Trust score — 8/10

/10 trustVerified

Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 133; Trending score 78/100 from researcher assessment.

More in the granite family

Sources