# Gemma 3 27B IT datasheet

> The 24 GB-card workhorse. Five of every six layers only look at the last 1,024 tokens, so the KV cache stays small as context grows.

Canonical: https://thelocalbench.com/models/gemma-3-27b · Data as of 2026-10-12 · The Local Bench (thelocalbench.com), data as of 2026-10-12, CC BY 4.0

Flags: [verified] checked against primary source · [sourced] cited · [estimate] formula · [sample] placeholder

## Specifications

| Field | Value | Flag | Source |
|---|---|---|---|
| params_total | 27.43 B | verified | https://huggingface.co/api/models/unsloth/gemma-3-27b-it |
| params_active | 27.43 B | verified | https://huggingface.co/api/models/unsloth/gemma-3-27b-it |
| layers | 62 | verified | https://huggingface.co/unsloth/gemma-3-27b-it/blob/main/config.json |
| kv_heads | 16 | verified | https://huggingface.co/unsloth/gemma-3-27b-it/blob/main/config.json |
| q_heads | 32 | verified | https://huggingface.co/unsloth/gemma-3-27b-it/blob/main/config.json |
| head_dim | 128 | verified | https://huggingface.co/unsloth/gemma-3-27b-it/blob/main/config.json |
| hidden_size | 5376 | verified | https://huggingface.co/unsloth/gemma-3-27b-it/blob/main/config.json |
| context | 131072 tokens | verified | https://huggingface.co/unsloth/gemma-3-27b-it/blob/main/config.json |
| sliding_window | 1024 tokens | verified | https://huggingface.co/unsloth/gemma-3-27b-it/blob/main/config.json |
| sliding_layers | 52 | verified | https://huggingface.co/unsloth/gemma-3-27b-it/blob/main/config.json |
| license | Gemma Terms of Use | verified | https://ai.google.dev/gemma/terms |

## Memory and quants (total = weights + F16 KV + overhead)

| Quant | bpw | File GB | At 8k | At 32k | At 128k | KLD | 24 GB card, 8k |
|---|---|---|---|---|---|---|---|
| Q3_K_M | 3.91 | 13.44 [verified] | 15.1 | 17.3 | 26.1 | 0.071 [sample] | fit |
| Q4_K_M | 4.85 | 16.55 [verified] | 18.2 | 20.4 | 29.2 | 0.017 [sample] | fit |
| Q5_K_M | 5.69 | 19.27 [verified] | 20.9 | 23.1 | 31.9 | 0.008 [sample] | fit |
| Q6_K | 6.56 | 22.17 [verified] | 23.8 | 26.0 | 34.8 | 0.004 [sample] | offload |
| Q8_0 | 8.5 | 28.71 [verified] | 30.4 | 32.6 | 41.4 | 0.001 [sample] | offload |
| F16 | 16 | 54.03 [verified] | 55.7 | 57.9 | 66.7 | — | offload |

## Speed (default quant, 8k context, single stream)

| Device | Verdict | Decode tok/s | Flag |
|---|---|---|---|
| [GeForce RTX 5060 Ti 16 GB](/hardware/rtx-5060-ti-16g.md) | offload | 7.1 | estimate |
| [GeForce RTX 4060 Ti 16 GB](/hardware/rtx-4060-ti-16g.md) | offload | 5.8 | estimate |
| [Radeon RX 7900 XTX 24 GB](/hardware/rx-7900-xtx-24g.md) | fit | 32.6 | estimate |
| [GeForce RTX 3090 24 GB (used)](/hardware/rtx-3090-24g-used.md) | fit | 31.8 | estimate |
| [CPU only · Ryzen 9 9950X + 128 GB DDR5](/hardware/cpu-ryzen-9-9950x-128g.md) | fit | 2.5 | estimate |
| [2× RTX 3090 (used) 48 GB](/hardware/2x-rtx-3090-48g-used.md) | fit | 31.8 | estimate |
| [Mac Studio M5 Max (base)](/hardware/mac-studio-m5-max.md) | fit | 11.7 | estimate |
| [Ryzen AI Max+ 395 mini PC 128 GB](/hardware/ryzen-ai-max-395-128g.md) | fit | 6.5 | estimate |
| [GeForce RTX 4090 24 GB](/hardware/rtx-4090-24g.md) | fit | 34.3 | estimate |
| [NVIDIA DGX Spark FE 128 GB](/hardware/dgx-spark-128g.md) | fit | 7.0 | estimate |
| [MacBook Pro M4 Max 40c · 128 GB](/hardware/macbook-pro-m4-max-128g.md) | fit | 13.9 | estimate |
| [Mac Studio M5 Ultra (base)](/hardware/mac-studio-m5-ultra.md) | fit | 30.6 | estimate |
| [GeForce RTX 5090 32 GB](/hardware/rtx-5090-32g.md) | fit | 60.9 | estimate |
| [Mac Studio M5 Ultra · 512 GB](/hardware/mac-studio-m5-ultra-512g.md) | fit | 30.6 | estimate |

## Quality

LMArena text score: 1357.7 [sourced] (LMArena text, overall, entry 'gemma-3-27b-it': 1357.7 (95% CI 1354.1–1361.4, 46,672 votes), leaderboard published 2026-10-08. CC-BY-4.0.)

JSON: https://thelocalbench.com/api/v1/models/gemma-3-27b.json · Configurator: https://thelocalbench.com/configurator?model=gemma-3-27b