One file of commented source code costs qwen3-coder-30b 210 tokens in English and 353 in German. That German figure (ouch, for a coding model we would love to run at home) is the second highest in our catalogue, 20% above the cheapest model measured — only granite-4.2-8b goes higher, at 381 — and it needs 17.28 GiB in Q4_K_M before any context is reserved.
qwen3-coder-30bModel sheet
granite-4.2-8bModel sheet
Now the twist we love: it is not alone up there. Three other models produce exactly 353, because all four share one tokenizer.
The same code, measured on 26 models
English is the baseline column, and the other three are measured against it. German is the column a Berlin team actually pays.
| Model | Size (reference quant) | English | German | vs the cheapest |
|---|---|---|---|---|
llama4-scout-17b |
60.87 GiB | 211 | 295 | — |
gpt-oss-20b |
10.83 GiB | 210 | 298 | +1% |
qwen3.8-27b |
15.33 GiB | 223 | 325 | +10% |
mistral-small-24b |
13.35 GiB | 217 | 327 | +11% |
gemma4-12b |
6.63 GiB | 239 | 332 | +13% |
flint-40b |
23.61 GiB | 210 | 335 | +14% |
glm-5.2 |
433.83 GiB | 209 | 343 | +16% |
deepseek-v4-flash |
127.28 GiB | 219 | 351 | +19% |
qwen3-8b |
4.68 GiB | 210 | 353 | +20% |
qwen3-coder-30b |
17.28 GiB | 210 | 353 | +20% |
granite-4.2-8b |
4.98 GiB | 238 | 381 | +29% |
Measured by Local AI Scope. Parallel corpus sha256[:16] 2f6c96b6ea0b161f, tokenized 2026-08-25 (granite-4.2-8b: 2026-09-24; flint-40b: 2026-10-23), add_special_tokens=False. Size is the published Q4_K_M artefact, or the nearest compression the author publishes (deepseek-v4-flash at IQ4_XS).
qwen3.8-27b comes from the same vendor but uses a newer tokenizer, at 248,320 tokens, and it lands 28 tokens lower on the same file.
qwen3.8-27bModel sheet
Of those four, qwen3-coder-30b is the only one built specifically for programming, and by far the largest: 17.28 GiB against 4.68 for qwen3-8b. A team choosing it pays for the coding specialisation and inherits the old tokenizer with it.
Shared tokenizers cross vendor lines, too. NVIDIA’s nemotron-3.5-super-120b and nemotron-3.5-lightning-30b return exactly the 217 and 327 of mistral-small-24b on this file, because all three run on the same tokenizer, Mistral’s Tekken, with 131,072 entries.
In English the difference nearly vanishes
The English column spans 209 to 239 tokens, a 14% spread, and qwen3-coder-30b sits near the efficient end of it at 210. Rank these models on English code alone and it looks excellent.
That is the transplant problem, and it is a bit like buying hiking boots after trying on only the left one. The spread that matters to a German team lives in the German column, and it is wider than the English one: 295 to 381 is 29%, against 14% in English.
What it costs in French and Spanish
| Language | qwen3-coder-30b |
Cheapest measured | Gap |
|---|---|---|---|
| English | 210 | 209 | +0% |
| Spanish | 269 | 252 (gpt-oss-20b) |
+7% |
| French | 309 | 277 (gpt-oss-20b) |
+12% |
| German | 353 | 295 (llama4-scout-17b) |
+20% |
Measured by Local AI Scope, same corpus and same file.
A French team pays a 12% surcharge on this model. A German team pays 20%, on a model that also asks for 17.28 GiB, when gpt-oss-20b sits just one percentage point off the best score at 10.83 GiB. Bigger and dearer in German: worth knowing before you pick it.
gpt-oss-20bModel sheet
What this does not say
To be straight with you: fewer tokens does not mean better code. We have measured what the same work costs on each tokenizer; we have not run these models or evaluated their output, because that needs a GPU and an evaluation protocol we do not have.
Calling a formula a benchmark is exactly the error this site exists to avoid, so the claim stops where the measurement stops.
The consequence for the machine, not just the invoice
Tokens are also how context memory is allocated, so the surcharge lands twice, like a big suitcase that costs extra at check-in and then eats your legroom. A repository excerpt that fits an 8k window in English needs roughly 13.8k in German on this tokenizer, and that extra context competes with the weights for the same memory.
qwen3-coder-30b in Q4_K_M is 17.28 GiB before any context, which rules it out on every 8 GB and 12 GB machine in our profiles and makes it a 24 GB proposition. And the effect shows up on small documents, not just long ones.
What to check before accepting a coding recommendation
One habit we would pass on to every team:
In this catalogue, models with vocabularies above 200,000 cluster between 1.29x and 1.33x for German when the four document types are taken together; those below 155,000 pay between 1.40x and 1.62x. Broken out by document type the two bands overlap, so treat this as a shortlist heuristic, not a per-document rule. qwen3-coder-30b sits at the far end of the second group, despite being a 2026 release.
Then check the fit for your actual machine and language rather than the headline parameter count, since the context has to fit next to the weights. The Model Finder does both, the token cost table has the per-language ratios for all 26 models, and the runtime profiles cover the program that will load it.