IBM released its reasoning-focused Granite 4.2 family under Apache 2.0 and made the weights available through Hugging Face. The models can use a thinking mode before answering, with a lower-computation setting for simpler questions. IBM lists vLLM, SGLang, OpenCode, Pi and OpenHands as supported runtimes. Its performance figures come from IBM's own evaluations.
IBM released its reasoning-focused Granite 4.2 family under the Apache 2.0 licence. The weights are available through Hugging Face; the report describes them as usable model weights rather than access limited to a closed service.[1]
The models can use a thinking mode before answering. A separate setting uses less computation for simpler questions. The announcement does not provide an independent benchmark for which work type requires which setting.[1]
IBM lists vLLM, SGLang, OpenCode, Pi and OpenHands among supported runtimes. The published performance figures come from IBM's own evaluations. The announcement does not present those results as an independent comparison. The Apache 2.0 licence and Hugging Face distribution define how the weights can be obtained, while the listed runtimes define the environments IBM names. Neither point changes the need to read IBM's performance figures as the company's own evaluation rather than independent verification. IBM released its reasoning-focused Granite 4.2 family on Hugging Face under Apache 2.0. The company offers thinking and lower-effort modes, while the announced performance results need to be read as IBM's own evaluations, not an independent comparison. This briefing keeps together the actor, reported scope and stated uncertainties of the event described by the source. It adds no separate development, definite outcome or comparison absent from that source; the headline isolates only this event.[1]
