Gemma 3 27B it vs Gemma 4 31B it

Gemma 4 31B it scores higher on the intelligence index, 15 against 5. Both take the same machine to start: the cheapest here that runs either is the Framework Desktop, 32GB, at $1,269. On it, Gemma 3 27B it is about 1.2× quicker: 9.8 tok/s against 8.3, both estimated from memory bandwidth. At 500k tokens a day the Framework Desktop, 32GB pays for itself in 103 years running Gemma 4 31B it, against 155 years running Gemma 3 27B it.

Run Gemma 3 27B it on the Framework Desktop, 32GB · or Gemma 4 31B it

Jump to: What the newer model changes · Side by side on the Framework Desktop, 32GB · How much use it takes to pay for the machine · Machines that run one and not the other · The assumptions behind both columns

What the newer model changes

Gemma 3 27B it is last generation. Gemma 4 31B it is the current dense Gemma model nearest it in size, 31.3B against 27.4B. On the intelligence index it scores 15 where Gemma 3 27B it scores 5. It asks more of the machine: 23 GB at 32k of context against 20 GB. The weights are 20 GB against 17 GB, and the cache at that window is 3.5 GB against 3.1 GB. 28 of the 37 machines priced here run it, against 30 for Gemma 3 27B it, and the cheapest is the same either way: the Framework Desktop, 32GB at $1,269.

Gemma 4 31B it takes 256k of context where Gemma 3 27B it stops at 128k, which is a ceiling rather than a setting: what you actually get is whatever the machine has room for.

Side by side on the Framework Desktop, 32GB

The cheapest machine that runs either model is the same one, so this is the pair doing the same work on the same hardware: Strix Halo Framework Desktop, 32GB, at $1,269.

Run Gemma 3 27B it on the Framework Desktop, 32GB · or Gemma 4 31B it

How much use it takes to pay for the machine

Everything above is at 500k tokens a day. Pay-back moves with how much you actually run, so here are both models at the five levels of use the calculator names, on the Framework Desktop, 32GB. Gemma 4 31B it pays for it sooner at every level of use, so which of them to run does not turn on how hard you work it.

The Framework Desktop, 32GB cannot generate 20M tokens a day on either model: at most 13.5M on Gemma 3 27B it and 11.5M on Gemma 4 31B it. Both figures on that row are for the most it can do.

Run Gemma 3 27B it at 20M tokens a day · or Gemma 4 31B it

Machines that run one and not the other

Gemma 3 27B it needs 20 GB of memory at 32k of context and Gemma 4 31B it needs 23 GB. That puts Gemma 3 27B it on 2 of the 37 machines priced here that Gemma 4 31B it does not, starting at $1,299.

Gemma 3 27B it is also head to head with Qwen3 8B above it on the leaderboard and Ministral 3 8B below it. Three more models need much the same memory: Qwen3.6 27B, Devstral Small 2 24B and Mistral Small 3.2 24B Instruct. Gemma 4 31B it is also head to head with Qwen3.5 122B-A10B above it on the leaderboard and Granite 4.2 30B below it. One more model needs much the same memory: Qwen3.6 35B-A3B.

The assumptions behind both columns

Both columns use the same usage: 500k tokens a day at 15:1 input to output, 32k of context, $0.17 per kWh, and today's API prices held flat. Speeds marked estimated are worked out from memory bandwidth rather than measured, and pay-back scales with them. Where nobody rents an open model by the token, its API prices are the nearest hosted model's, named beside them. Machines are the 37 here with a published price that are still sold. Change any of it in the calculator.

More head to head: every machine that runs Gemma 3 27B it · every machine that runs Gemma 4 31B it · every other match-up · both against the frontier · the quickest pay-back at each level of use