Gemma 4 E4B vs Devstral 2 123B
Gemma 4 E4B and Devstral 2 123B both score 9 on the intelligence index. The cheapest machine here that runs Devstral 2 123B is the Framework Desktop, 128GB, at $3,449. Gemma 4 E4B runs on the Mac mini M6, 16GB at $899, $2,550 less. On the Framework Desktop, 128GB, the cheapest machine here that runs both, Gemma 4 E4B is about 18.6× quicker: 41 tok/s against 2.2, both estimated from memory bandwidth. At 500k tokens a day the Framework Desktop, 128GB pays for itself in 59 years running Devstral 2 123B, against 1,468 years running Gemma 4 E4B.
| Gemma 4 E4B | Devstral 2 123B | |
|---|---|---|
| Intelligence index | 9 | 9 |
| Class | Below every hosted tier | Below every hosted tier |
| Weights | 5.2 GB | 75 GB |
| Needs at 32k | 5.7 GB | 87 GB |
| Quantisation | QAT Q4_0 | Q4_K_M |
| Parameters | 8B (4.5B active) | 125B |
| Max context | 128k | 256k |
| API price per 1M | $0.018 in / $0.09 outpriced as gpt-oss-20b | $0.4 in / $2 out |
| Licence | Apache 2.0 | other (see repo) |
| Machines here that run it | 37 of 37 | 12 of 37 |
| Cheapest machine that runs it | Mac mini M6, 16GB $899 | Strix Halo Framework Desktop, 128GB $3,449 |
Run Gemma 4 E4B on the Mac mini M6, 16GB · or Devstral 2 123B on the Framework Desktop, 128GB
Jump to: Side by side on the Framework Desktop, 128GB · How much use it takes to pay for the machine · Machines that run one and not the other · The assumptions behind both columns
Side by side on the Framework Desktop, 128GB
The table above gives each model the cheapest machine that runs it, and those are two different machines, so nothing in it is a like-for-like race. The Strix Halo Framework Desktop, 128GB is the cheapest machine here that runs both, so this is the pair doing the same work on the same hardware.
| Gemma 4 E4B | Devstral 2 123B | |
|---|---|---|
| Speed at 32k | 41 tok/s estimated | 2.2 tok/s estimated |
| Pay-back on this machine | Pays back in 1,468 years | Pays back in 59 years |
| API cost per month | $0.34priced as gpt-oss-20b | $7.61 |
Run Gemma 4 E4B on the Framework Desktop, 128GB · or Devstral 2 123B
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, 128GB. Devstral 2 123B pays for it sooner at every level of use, so which of them to run does not turn on how hard you work it.
| A day's use | Gemma 4 E4B | Devstral 2 123B |
|---|---|---|
| 50ka few chats a day | 14,676 years | 585 years |
| 200klight assistant use | 3,669 years | 146 years |
| 1Ma moderate coding-assistant day | 734 years | 29 years |
| 4Mheavy coding with an agent | 183 years | 9.6 yearsits ceiling |
| 20Magents running most of the day | 37 years | 9.6 yearsits ceiling |
The Framework Desktop, 128GB generates at most 3.06M tokens a day on Devstral 2 123B, so that column's figures from 4M tokens a day up are for the most it can do, not for the whole of what was asked.
Run Gemma 4 E4B at 20M tokens a day · or Devstral 2 123B
Machines that run one and not the other
Gemma 4 E4B needs 5.7 GB of memory at 32k of context and Devstral 2 123B needs 87 GB. That puts Gemma 4 E4B on 25 of the 37 machines priced here that Devstral 2 123B does not, starting at $899.
| Machine | Price | Memory | Speed on Gemma 4 E4B | Pay-back |
|---|---|---|---|---|
| Mac mini M6, 16GB | $899 | 16 GB | 13 tok/s estimated | Pays back in 611 years |
| Mac mini M6, 24GB | $1,099 | 24 GB | 15 tok/s estimated | Pays back in 633 years |
| Strix Halo Framework Desktop, 32GB | $1,269 | 32 GB | 41 tok/s estimated | Pays back in 540 years |
| Mac mini M6, 32GB | $1,299 | 32 GB | 15 tok/s estimated | Pays back in 748 years |
| MacBook Air M5 (13-inch), 16GB | $1,299 | 16 GB | 13 tok/s estimated | Pays back in 882 years |
| AMD Radeon AI PRO R9700, 32GB | $1,299card only | 32 GB | 96 tok/s estimated | Pays back in 534 years |
19 more, on the Gemma 4 E4B page.
Gemma 4 E4B is also head to head with gpt-oss-20b above it on the leaderboard. Devstral 2 123B is also head to head with DeepSeek-R1-Distill-Qwen-32B below it on the leaderboard. One more model needs much the same memory: GLM-4.5-Air.
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 4 E4B · every machine that runs Devstral 2 123B · every other match-up · both against the frontier · the quickest pay-back at each level of use