Granite 4.2 30B
IBM Granite 4.2 30B is a dense reasoning model with switchable thinking modes, reasoning-augmented tool calling, 512K context, and an Apache 2.0 license.
Model Overview
Capabilities, design details, and architectural traits
Granite 4.2 30B - IBM's flagship dense reasoning model with switchable thinking
Granite 4.2 30B is the largest variant of IBM's dense reasoning model family, built for complex reasoning, coding, and agentic tool-calling workloads. Its defining feature is native chain-of-thought reasoning inside `` tags, combined with the ability to turn that reasoning on, off, or dial it down per query.
What sets Granite 4.2 30B apart
| Trait | Detail |
|---|---|
| Flexible thinking modes | Switch between thinking (default), non-thinking, and low-effort modes within a single model to balance depth against latency per query |
| Reasoning-augmented tool calling | Reasons about which tools to invoke and why before making the call, for more accurate function calls in agentic workflows |
| Extended long context | 128K context window natively, with a long-context extension to 512K available only on the 30B model |
| Open with enterprise assurances | Apache 2.0 license with cryptographic signatures, ISO certification, and full transparency disclosures |
| Agentic RL training | A specialized agentic RL phase covering software engineering, terminal-based coding, and search-driven workflows, on top of foundational RL and RLHF alignment |
| Synthetic code training | Trained on 1 trillion tokens of synthetic code from IBM's CodeAlchemy pipeline, plus a mid-training step to unlock more reasoning power |
| Built-in speculative decoding | A speculative decoding layer speeds up text output while serving more users |
Built for enterprise agentic work
IBM positions Granite 4.2 as purpose-built for agentic workflows where tasks are ambiguous and multi-step. The 30B model serves as the flagship for deeper reasoning and complex coding, while smaller siblings (3B, 8B) handle high-throughput agentic tasks. Software engineering agents built on it can navigate codebases, handle multi-step development tasks, and operate in terminal environments. It is tested across 12 languages, including English, German, Japanese, Arabic, Korean, and Chinese.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | IBM | Anthropic | Anthropic |
| Release Date | August 25, 2026 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | - | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 131K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | $0.16 Best Input Pricing | $4 | $2 |
| Output Pricing | $0.65 Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | text | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 12.7 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | 29.9 | - | - |
| Agentic Index | 3.5 | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from IBM
Other models by IBM
Top AI Models
Leading alternatives by intelligence score
Claude Opus 5.5
Muse Spark 1.3
GPT-6 Astra
