All models › google

Gemma 4 26B A4B

google · Apache 2.0 license · on Arena since 31 Mar 2026

Free API Lab chat Download
Arena score1434#90 of 416 in Chat, paid models included
Size25.8 B4 B active per token (mixture of experts)
4-bit download16 GBweights only, Q4_K_M
Max context256Ktokens, from the model config

Where to use it for free

Arena scores

ArenaScoreRankVotes
Chat1434#90 of 4166K
Coding1359#109 of 1442K
Vision1260#55 of 16524K

Rank among all models, paid ones included. Source: Arena leaderboard dataset.

Speed on popular hardware

HardwareGPU memoryFits?Speed
RTX 3060 12 GB12 GBGPU + 64 GB RAM42 tok/s
RTX 4060 Ti 16 GB16 GBGPU + 64 GB RAM46 tok/s
RTX 309024 GBFits in GPU165 tok/s
RTX 409024 GBFits in GPU177 tok/s
RTX 509032 GBFits in GPU316 tok/s
RX 7900 XTX24 GBFits in GPU169 tok/s
2 × RTX 309048 GBFits in GPU160 tok/s
4 × RTX 309096 GBFits in GPU151 tok/s
Mac M4 Max 128 GB96 GBFits in GPU96 tok/s
Ryzen AI Max+ 395 128 GB96 GBFits in GPU45 tok/s
8 × H100 SXM640 GBFits in GPU487 tok/s

Estimate at 4-bit with 32K context, 64 GB of DDR5 for offload. Speed ≈ memory bandwidth × 70% ÷ data read per token; real numbers vary by about ±30%.

GPU memory needed

Level8K context32K context128K contextQuality vs full
Q3_K_M14 GB15 GB19 GB+7.4% noticeable loss
Q4_K_M17 GB18 GB22 GB+1.9% small loss
Q5_K_M19 GB20 GB25 GB+0.6% very small loss
Q6_K22 GB23 GB27 GB+0.2% near-lossless
Q8_029 GB30 GB34 GB≈ 0% same as full
F1653 GB54 GB58 GB≈ 0% same as full

Weights + context cache (from the model's architecture) + runtime. Quality: increase in perplexity against the full model, from llama.cpp tests, scaled by model size. Simulate your own setup.

Chat score over time

Gemma 4 26B A4B 1434Best paid model

Arena score from 17 Apr 2026 to 15 Oct 2026.

Vision score over time

Gemma 4 26B A4B 1260Best paid model

Arena score from 10 Apr 2026 to 9 Oct 2026.

Similar free models in Chat