Gemma 4 12B it AEON Abliterated K4 BF16
Fine-tune of Gemma 4 12B IT
composite n/acompare to base
fine-tune not yet benchmarked
Measured axis deltas appear after both rows have board data.
Gemma 4 12B it AEON Abliterated K4 BF16
Every measured variant of this model: how much quality each quant keeps, and the VRAM and speed it costs to run.
VRAM footprint vs Local Intelligence Index
LB-2026-07.2 | 25/22.5/22.5/22.5/7.5. Where this model and current-lane family runs land vs the frontier anchors.
Swipe horizontally to inspect the full chart →
Base modelDashed vertical lines mark common VRAM tiers.Amber points are synthetic demo preview data.
Variant profiles
Complete rows are ordered by Local Intelligence Index; partial rows show their measured axes but are not ranked. The VRAM/Fits columns (8Kcontext) tell you what your card needs. Ranks are within this family's variants.
Column guide
Variant — quant label plus, where sha-verified, the publisher of the exact weights file benchmarked; rows without a source line predate artifact identity records.
VRAM @8k — model weights, KV cache, and runtime headroom at 8k context.
Fits — the smallest common GPU VRAM tier above that estimate.
Prefill tok/s — prompt-processing speed.
Decode tok/s — generated-token speed after the prompt.
Overall tok/s — completion throughput across the full benchmark.
File size — the benchmarked model artifact on disk.
Runtime — the serving engine and version.
Run — the immutable benchmark receipt. Live rows link to the public submission record until the run is baked into the static site.
Pick the largest quant whose VRAM @8k fits your card.
Swipe horizontally for all variant metrics →
| Rank (this family) | Variant | Local Intelligence IndexLB-2026-07.2 | 25/22.5/22.5/22.5/7.5 | Agentic | Knowledge | Instruction | Coding | Math | VRAM @8k | Fits | Prefill tok/s | Decode tok/s | Overall tok/s | File size | Runtime | Run |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 40.3±2.7 Agentic 4.2 / Knowledge 75.6 / Instruction 66.7 / Coding 28.4 / Math 11.5 | 4.2±5.8 | 75.6±4.8 | 66.7±5.4 | 28.4±7.8 | 11.5±5.8 | 6.7 GB footprint | n/a | 5,383.5 | 147.5 | 137.1 | — | llama.cppb9852/fd1a05791 | receipt | |
| — | Q8_0by mradermacher | no run yet | — | — | — | — | — | 14.3 GB | 16 GB | — | 12.7 GB | — | benchmark it | ||
| — | Q6_Kby mradermacher | no run yet | — | — | — | — | — | 11.4 GB | 12 GB | — | 9.8 GB | — | benchmark it | ||
| — | Q5_K_Mby mradermacher | no run yet | — | — | — | — | — | 10.1 GB | 12 GB | — | 8.5 GB | — | benchmark it | ||
| — | Q4_K_Mby mradermacher | no run yet | — | — | — | — | — | 9 GB | 12 GB | — | 7.4 GB | — | benchmark it | ||
| — | Q3_K_Mby mradermacher | no run yet | — | — | — | — | — | 7.7 GB | 8 GB | — | 6.1 GB | — | benchmark it | ||
| — | Q2_Kby mradermacher | no run yet | — | — | — | — | — | 6.4 GB | 8 GB | — | 4.8 GB | — | benchmark it |
vs base