# IBM Granite

37 Hacker News commenters recommended IBM Granite and 18 criticized it, November 2025 to November 2026. It was recommended in 11 of 13 months; owners who said how long report a median of 0 years (3 owners). It meets the Hacker Approved badge rule.

**Verdict:** Small open-weight models from IBM that people run locally, praised for speed, efficiency and strong results for their size. Critics say they trail rival models on benchmarks and lack reasoning training.

## Why people recommend it

- Fast and efficient on local hardware (5 comments): "This makes then very token efficient and particularly well suited for local inference" ([source](https://news.ycombinator.com/item?id=47962619)) "I found these to be a great balance of speed and consistency on local hardware" ([source](https://news.ycombinator.com/item?id=46450492)) "The 4b they released was not good for my needs" ([source](https://news.ycombinator.com/item?id=47960682))
- Capable for its small size (5 comments): "genuinely not bad for a tiny 3B model - it can at least output valid SVG" ([source](https://news.ycombinator.com/item?id=47964402)) "a Granite model at nearly half the size has no problems" ([source](https://news.ycombinator.com/item?id=45924984)) "super good small parameter lineup for local tasks" ([source](https://news.ycombinator.com/item?id=50293033))
- Useful for agentic and tool-calling work (5 comments): "I've found them to be quite capable in content retrieval and long-running workflows" ([source](https://news.ycombinator.com/item?id=50239602)) "For coding I use qwen3.6:35b-mlx and agentic granite4.1:3b works fine." ([source](https://news.ycombinator.com/item?id=48565964)) "granite4.2 is optimised for agentic work and only needs 5GB of memory." ([source](https://news.ycombinator.com/item?id=49444893))

## What people warn about

- Trails rival models on benchmarks (6 comments): "qwen3.5 9b outperforms granite 4.1 30b by a huge amount (32 vs 15 on artificialanalysis benchmark)" ([source](https://news.ycombinator.com/item?id=47961399)) "The Granite models have always been at least one year behind in terms of capabilities." ([source](https://news.ycombinator.com/item?id=47985931)) "Their granite series appears to be nowhere close to competitive." ([source](https://news.ycombinator.com/item?id=48570117))
- Lacks reasoning training (2 comments): "It's strange that they don't include reasoning training (RLVR)." ([source](https://news.ycombinator.com/item?id=47961924)) "it didn't support reasoning which limited its performance" ([source](https://news.ycombinator.com/item?id=49101715))
- Poor output on some tasks (4 comments): "failed completely because it was just writing non-descriptive massages" ([source](https://news.ycombinator.com/item?id=48367109)) "i just got Granite4.1 tell me about YOLO repeatedly for no reasons" ([source](https://news.ycombinator.com/item?id=47993643)) "Its hopeless for basic task like parsing/generating headers" ([source](https://news.ycombinator.com/item?id=47970752))
- Clinical, wordy tone (2 comments): "Quick vibe check of it- 8B @ Q6 - seems promising. Bit of a clinical tone" ([source](https://news.ycombinator.com/item?id=47960689)) "Reams and reams of corporatese about why it wasn’t wrong when it was." ([source](https://news.ycombinator.com/item?id=50246781))

## Alternatives hackers recommend

- [Qwen 3.8 27B](https://hackerapproved.com/p/qwen-3-8-27b.md): 1173 recommend
- [Z.ai GLM](https://hackerapproved.com/p/glm-5-2.md): 1043 recommend
- [Kimi](https://hackerapproved.com/p/kimi-k3.md): 931 recommend
- [Gemma 4](https://hackerapproved.com/p/gemma-4.md): 713 recommend
- [Qwen](https://hackerapproved.com/p/qwen.md): 474 recommend

## Cite as

Hacker Approved, "IBM Granite", https://hackerapproved.com/p/granite-4-1-8b/ (counts from 1,795,640 Hacker News comments, November 2025 to November 2026).