Gemma 4 26B A4B
Google's open-weight mixture-of-experts language model, 26B total and about 4B active parameters.
HN commenters like it as a fast local model on Macs and consumer GPUs1234 for writing help, language study and simpler coding tasks.5678 The most common caveat is that it is weak at unsupervised coding910 and needs a human or a tight harness to keep it on track.1112
Why people recommend it
Simpler coding help · 3
Used for code review, text2sql, snippets and co-pilot style work on small to medium tasks.
“For simpler programming related (e.g. text2sql) as well as synthetic data generation, current best for me is llama.cpp + Gemma-4-26B-A4B”
“it's fast and good enough that it could handle being a "co-pilot" on small to medium context tasks”
Show 1 more
“It's a very good model when the harness keeps it on rails.”
Everyday writing tasks · 2
Handles spell checking, emails, tickets, message tone review and text correction.
“I find them quite useful!”
“Gemma 4 26B A4B gives me steady 20 tps.”
Speed · 3
Commenters describe it as fast, with one reporting a steady 20 tokens per second on an M4 Pro.
“Gemma 4 26B A4B gives me steady 20 tps.”
“I have switched to gemma-4-26b-a4b. At least in benchmarks, it scores higher and is faster.”
Show 1 more
“it's fast and good enough that it could handle being a "co-pilot" on small to medium context tasks”
Learning and questions · 2
Used for language study and learning Rust, with results described as rather good.
“`gemma-4-26b-a4b` is suprisingly capable”
“it's fast and good enough that it could handle being a "co-pilot" on small to medium context tasks”
Runs on local hardware · 2
Offers a good balance for small local setups and is preferred by some over a similar Qwen model.
“I find that Gemma 4 26b a4b gives the best balance here”
“though I use Gemma 4 26B-A4B more than Qwen3.6 35B-A3B”
Simple architecture · 1
One engine builder chose it over Qwen for its simpler architecture and kernels.
“I switched to Gemma because of its simpler architecture and kernels”
Caveats and dissent
Unreliable coding · 2
Commenters report it is not good for coding on its own and should not be trusted to one-shot changes.
“I used Gemma4 26B before and it wasn't good in my testing.”
“it's fast and good enough that it could handle being a "co-pilot" on small to medium context tasks”
Needs a harness · 1
One user says it is very good only when a harness keeps it on rails.
“It's a very good model when the harness keeps it on rails.”
Stronger alternatives · 1
One commenter recommends Qwen3.6 models first and suggests Gemma only if it suits your tasks better.
“(or gemma 4 26B if gemma models perform better for your tasks)”
Often compared with
1 moved from Qwen 3.6 to Gemma 4 26B A4B
1 moved from Gemma 4 26B A4B to Qwen 3.5
1 moved from Gemini 2.5 Flash to Gemma 4 26B A4B
Who recommends it in their own words
Threads where it came up
- Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding? 3 mentions
- LFM2.5-24B-A2B: Built for fast on-device agents 1 mention
- Best LLM for every budget, updated daily 1 mention
- Why your local LLM feels dumber than it is 1 mention
- Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac 1 mention
- Self-hosting Kimi K3: 20% more hardware cost, 20% better task resolution 1 mention
Also written as Gemma 4 26B, Gemma 4 26B-A4B, Gemma-4-26B-A4B. Data: gemma-4-26b-a4b.json. Quotes are verbatim; the verdict and theme labels are written by a model from these comments, and every clause links to its sources.
All comments 12
“I used Gemma4 26B before and it wasn't good in my testing.”
“I find them quite useful!”
“Gemma 4 26B A4B gives me steady 20 tps.”
“I switched to Gemma because of its simpler architecture and kernels”
“`gemma-4-26b-a4b` is suprisingly capable”
“I have switched to gemma-4-26b-a4b. At least in benchmarks, it scores higher and is faster.”
“I find that Gemma 4 26b a4b gives the best balance here”
“For simpler programming related (e.g. text2sql) as well as synthetic data generation, current best for me is llama.cpp + Gemma-4-26B-A4B”
“It's a very good model when the harness keeps it on rails.”
“it's fast and good enough that it could handle being a "co-pilot" on small to medium context tasks”
“though I use Gemma 4 26B-A4B more than Qwen3.6 35B-A3B”
“(or gemma 4 26B if gemma models perform better for your tasks)”