Home›Data changelog

Data changelog

What changed in the data, and the day it changed

Model catalogues go stale fast. This page is the receipt: every model or machine that comes in, every figure we correct, dated and linked to its own sheet.

This log Computed from the data
Changes recorded97
Days with changes38
Models measured2026-09-30
Machine data2026-10-23
Models today26
Machines today18
DateWhat changedWhere
4 runtime claims re-checkedllama-cpp ollama vllm
2 runtime versions updatedllama-cpp v0.7.0 / b11681 → v0.7.0 / b11850 ollama 0.40.9 → 0.40.10
5 runtime claims re-checkedvllm
Provenance re-checked on 2 figuresryzen-aimax395-128 ryzen-aimax495-192
2 runtime claims re-checkedollama
1 model addedflint-40b
1 runtime version updatedollama 0.40.8 → 0.40.9
4 runtime claims re-checkedollama vllm
2 runtime claims re-checkedllama-cpp
1 runtime version updatedollama 0.40.7 → 0.40.8
8 runtime claims re-checkedlm-studio ollama vllm
2 runtime claims re-checkedllama-cpp ollama
2 runtime versions updatedllama-cpp v0.7.0 / b11681 → v0.7.0 / b11681 ollama 0.40.6 → 0.40.7
1 runtime version updatedvllm 0.31.0 → 0.32.0
1 model addednemotron-3.5-lightning-30b
1 runtime claim re-checkedollama
1 runtime version updatedollama 0.40.5 → 0.40.6
2 runtime claims re-checkedmlx
1 model addednemotron-3.5-super-120b
1 runtime claim re-checkedmlx
1 runtime version updatedmlx mlx-lm 0.32.0 (mlx 0.32.4) → mlx-lm 0.32.2 (mlx 0.32.4)
5 runtime claims re-checkedllama-cpp ollama
2 runtime versions updatedllama-cpp v0.6.0 / b11515 → v0.7.0 / b11681 ollama 0.40.4 → 0.40.5
Context memory recomputed on 5 modelsgemma4-12b gemma4-26b-a4b gemma4-31b gemma4-e2b gemma4-e4b
Provenance re-checked on 5 figuresgemma4-12b gemma4-26b-a4b gemma4-31b gemma4-e2b gemma4-e4b
2 runtime versions updatedmlx mlx-lm 0.32.0 (mlx 0.32.3) → mlx-lm 0.32.0 (mlx 0.32.4) ollama 0.40.3 → 0.40.4
1 runtime claim re-checkedollama
1 runtime claim re-checkedllama-cpp
1 runtime version updatedlm-studio 0.4.26-4 → 0.4.26-4
1 runtime claim re-checkedollama
1 runtime version updatedollama 0.40.2 → 0.40.3
1 model addedqwen3.5-9b
7 runtime claims re-checkedollama
6 runtime claims re-checkedollama
3 runtime versions updatedllama-cpp v0.6.0 / b11491 → v0.6.0 / b11515 lm-studio 0.4.25-1 → 0.4.26-4 ollama 0.40.1 → 0.40.2
2 runtime claims re-checkedlm-studio
1 price updatedryzen-aimax495-192 6599 USD → 6799 USD (US)
Provenance re-checked on 1 figureryzen-aimax395-128
2 runtime versions updatedllama-cpp v0.6.0 / b11435 → v0.6.0 / b11491 ollama 0.40.0 → 0.40.1
1 runtime claim re-checkedollama
5 runtime claims re-checkedlm-studio ollama
3 runtime versions updatedlm-studio 0.4.25-1 → 0.4.25-1 ollama 0.35.1 → 0.40.0 vllm 0.31.0 → 0.31.0
Context memory recomputed on 1 modelllama4-scout-17b
1 model addedmuse-glimmer-30b
Provenance re-checked on 1 figureglm-5.3-flash
3 runtime claims re-checkedllama-cpp
1 runtime version updatedllama-cpp v0.5.0 / b11413 → v0.6.0 / b11435
2 runtime claims re-checkedllama-cpp vllm
2 runtime versions updatedllama-cpp v0.5.0 / b11370 → v0.5.0 / b11413 vllm 0.30.0 → 0.31.0
1 model addedglm-5.3-flash
Context memory recomputed on 1 modelglm-5.2
Provenance re-checked on 1 figureglm-5.2
Architecture corrected on 2 modelsdeepseek-v4-flash glm-5.2
2 runtime versions updatedllama-cpp v0.5.0 / b11321 → v0.5.0 / b11370 ollama 0.35.0 → 0.35.1
3 runtime claims re-checkedllama-cpp
1 runtime version updatedllama-cpp v0.5.0 / b11312 → v0.5.0 / b11321
5 runtime claims re-checkedllama-cpp
2 machines addedryzen-aimax395-128 ryzen-aimax495-192
2 runtime claims re-checkedmlx
2 runtime versions updatedllama-cpp v0.5.0 / b11272 → v0.5.0 / b11312 mlx mlx-lm 0.31.3 (mlx 0.32.3) → mlx-lm 0.32.0 (mlx 0.32.3)
2 runtime versions updatedllama-cpp v0.5.0 / b11246 → v0.5.0 / b11272 ollama 0.34.4 → 0.35.0
License corrected on 1 modelllama4-scout-17b
1 model addedqwen3.8-flash-next
2 runtime versions updatedllama-cpp v0.5.0 / b11160 → v0.5.0 / b11246 mlx mlx-lm 0.31.3 (mlx 0.32.2) → mlx-lm 0.31.3 (mlx 0.32.3)
1 runtime claim re-checkedvllm
2 runtime claims re-checkedvllm
9 runtime claims re-checkedlm-studio mlx vllm
2 runtime claims re-checkedlm-studio
6 runtime claims re-checkedollama
2 runtime versions updatedllama-cpp v0.4.1 / b11119 → v0.5.0 / b11160 ollama 0.34.3 → 0.34.4
1 model addedgranite-4.2-8b
5 machine figures correctedmac-mini-m5pro-64 mac-mini-m6-16 mac-mini-m6-32 mac-studio-m5max-128 mac-studio-m5ultra-512
5 runtime claims re-checkedllama-cpp lm-studio mlx ollama vllm
3 prices updatedmac-mini-m5pro-64 1699 USD → 2899 USD (US) mac-mini-m6-32 899 USD → 1299 USD (US) mac-studio-m5max-128 2499 USD → 5099 USD (US)
Provenance re-checked on 2 figuresmac-mini-m6-16 mac-studio-m5ultra-512
2 runtime versions updatedllama-cpp v0.4.1 / b11095 → v0.4.1 / b11119 ollama 0.34.2 → 0.34.3
5 runtime versions updatedllama-cpp v0.4.1 / b11028 → v0.4.1 / b11095 lm-studio 0.4.24-1 → 0.4.25-1 mlx mlx-lm 0.31.3 (mlx 0.32.2) → mlx-lm 0.31.3 (mlx 0.32.2) ollama 0.34.2 → 0.34.2 vllm 0.29.0 → 0.30.0
4 machine figures correctedmac-mini-m5pro-64 mac-mini-m6-32 mac-studio-m5max-128 mac-studio-m5ultra-512
1 machine addedmac-mini-m6-16
2 machine figures correctedmac-m4max-64 mac-m4pro-24
1 machine addedmac-mini-m5pro-64
1 price addedmac-m4-16 new → 1499 USD (US)
5 runtime versions updatedllama-cpp v0.4.1 / b11026 → v0.4.1 / b11028 lm-studio 0.4.24-1 → 0.4.24-1 mlx mlx-lm 0.31.3 (mlx 0.32.2) → mlx-lm 0.31.3 (mlx 0.32.2) ollama 0.34.1 → 0.34.2 vllm 0.29.0 → 0.29.0
1 machine addedmac-studio-m5max-128
5 runtime versions updatedllama-cpp v0.2.0 / b10618 → v0.4.1 / b11026 lm-studio 0.4.21-2 → 0.4.24-1 mlx mlx-lm 0.31.3 (mlx 0.32.1) → mlx-lm 0.31.3 (mlx 0.32.2) ollama 0.32.15 → 0.34.1 vllm 0.27.1 → 0.29.0
1 machine addedmac-studio-m5ultra-512
1 price addedmac-mini-m6-32 new → 899 USD (US)
2 machines addedmac-mini-m6-32 xiaomi-ai-cube-80
216 language-cost figures re-measureddeepseek-v4-flash gemma4-12b gemma4-26b-a4b gemma4-31b gemma4-e2b gemma4-e4b +12 more
Provenance re-checked on 40 figuresmac-m4-16 mac-m4max-64 mac-m4pro-24 minipc-ai-32 portatil-cpu-16 rtx2x3090-48 +4 more
10 names correctedmac-m4-16 mac-m4max-64 mac-m4pro-24 minipc-ai-32 portatil-cpu-16 rtx2x3090-48 +4 more
9 models addeddeepseek-v4-flash gemma4-26b-a4b gemma4-31b gemma4-e2b glm-5.2 gpt-oss-120b +3 more
6 file sizes correcteddeepseek-v4-flash Q8_0 10.15 GiB → withdrawn gemma4-12b Q8_0 12.23 GiB → 11.80 GiB gemma4-26b-a4b Q8_0 25.45 GiB → 25.02 GiB gemma4-31b Q8_0 30.87 GiB → 30.39 GiB gemma4-e2b Q8_0 4.79 GiB → 4.70 GiB +1 more
2 file sizes correctedgemma4-12b Q8_0 11.80 GiB → 12.23 GiB gemma4-e4b Q8_0 7.63 GiB → 7.72 GiB
Catalogue opened with 10 machinesmac-m4-16 mac-m4max-64 mac-m4pro-24 minipc-ai-32 portatil-cpu-16 rtx2x3090-48 +4 more
Catalogue opened with 9 modelsgemma4-12b gemma4-e4b gpt-oss-20b qwen3-1.7b qwen3-4b-2507 qwen3-8b +3 more
Catalogue opened with 5 runtimesllama-cpp lm-studio mlx ollama vllm

This page is not written by hand: every line is computed by comparing versions of the dataset itself, so it cannot report a change that did not happen — nor stay silent about one that did.

How this page is made

Computed from the data, not written about it

Nobody types an entry here. A generator compares the dataset this site serves against every earlier version of it and reports the difference: models in and out, file sizes corrected, language costs re-measured. If the data changes and no one writes a word, this page changes anyway; if nothing changed, it stays quiet. That is the whole point — a log kept by hand stops being kept, and a dated lie is worse than no log at all.

Why each figure is measured the way it is

A changelog tells you the data moves. Whether it moves for your machine, at the context your work needs and in your language, is what the Model Finder answers.

Check it against my machine