Model datasheet · Gemma 4

Gemma 4 26B-A4B

Google's answer to the fast-MoE era: 26B of knowledge at 4B speed, vision, 256K window, QAT 4-bit weights, and finally an Apache license. The strongest all-rounder in the 16GB class right now.

Vendor: Google
Architecture: MoE · 25.2B total / ~3.8B active
Context: 262,144 tokens
License: Apache-2.0
Released: 2026-07-02
Vision: Yes

→ GGUF (llama.cpp / LM Studio / Ollama) → MLX (Apple Silicon)

§1 Characteristics by quantization band

Q2–Q3 — 2–3 bit

13GB weights · e.g. Q3_K_M + KV: 576MB at 8K · 1.7GB at 32K · 6.2GB at 128K

— Not yet rated · contributions welcome —

Q4–Q5 — 4–5 bit

14GB weights · e.g. Q4_0 , Q4_K_M , mlx-4bit + KV: 576MB at 8K · 1.7GB at 32K · 6.2GB at 128K

Math & reasoning

7/10

data checked Aug 2026

quant-degradation-community — community consensus @ Q4_0 · aggregated · 2026-08-30
Editorial derivation: QAT Q4_0 is the intended local build; small loss vs full.

Coding

7/10

data checked Aug 2026

quant-degradation-community — community consensus @ Q4_0 · aggregated · 2026-08-30
Editorial derivation: QAT limits quant loss.

Tool calling / agentic

6/10

data checked Aug 2026

quant-degradation-community — community consensus @ Q4_0 · aggregated · 2026-08-30
Editorial derivation: MoE expert weights are quant-sensitive; slight downgrade at 4-bit.

Q6–Q8 — 6–8 bit

23GB weights · e.g. Q6_K + KV: 576MB at 8K · 1.7GB at 32K · 6.2GB at 128K

— Not yet rated · contributions welcome —

FP16 — full precision (fp16/bf16)

50GB weights · e.g. bf16 + KV: 576MB at 8K · 1.7GB at 32K · 6.2GB at 128K

Math & reasoning

8/10

data checked Aug 2026

vendor-model-card — AIME 2026 (no tools) = 88.3 @ bf16 · vendor-reported · 2026-08-30
GPQA Diamond: 82.3, MMLU-Pro: 82.6. Too new for independent leaderboards (absent from Vectara HHEM and Aider).

Coding

8/10

data checked Aug 2026

vendor-model-card — LiveCodeBench v6 = 77.1 @ bf16 · vendor-reported · 2026-08-30
A generational leap over Gemma 3's coding — but vendor-reported only so far.

Tool calling / agentic

7/10

data checked Aug 2026

vendor-model-card — Tau2 average = 68.2 @ bf16 · vendor-reported · 2026-08-30
Gemma 4 adds trained-in tool calling — fixing the family's old gap.

Long context

4/10

data checked Aug 2026

vendor-model-card — MRCR v2 @128K = 44.1 @ bf16 · vendor-reported · 2026-08-30
The 256K window is real but retrieval quality drops well before it fills — the vendor's own number says so.

§2 Known issues & what fixes them

ISSUE-01 moderate Q2–Q3 / Q4–Q5 / Q6–Q8 / FP16 · long-context

The 256K window oversells it: retrieval accuracy at 128K is only 44.1% (vendor's own MRCR v2 number). Don't stuff the window and expect recall.

Fix

RAG (retrieval) [strong]

Retrieve narrowly instead of filling the context — smaller, relevant chunks beat a full window.

Evidence · 1 source · anecdotal

Gemma 4 model card (MRCR v2 128K: 44.1) (vendor, 2026-08-30)

ISSUE-02 moderate Q2–Q3 / Q4–Q5 / Q6–Q8 / FP16 · factuality

Early reports of the Gemma-family habit: hallucinated names and specific details on general-knowledge questions. No independent hallucination benchmark lists it yet.

Fix

RAG (retrieval) [strong]

Ground factual work until independent measurements arrive.

Evidence · 1 source · anecdotal

Gemma 4 factual hallucination reports (other, 2026-08-30)