Best Local LLMs for Translation

Written by Jakub Rusinowski · Last updated June 26, 2026

Translating between languages and working in a language other than English.

Top pick: GLM-4.7 9B

Scores 96.1/100 for translation and multilingual work. 9B parameters, needing about 6.2 GB at Q4_K_M, 125K context, Apache-2.0.

Ranked for translation and multilingual work

ModelScoreParamsContextLicenceQuality index
1. GLM-4.7 9B96.19B125KApache-2.0— (estimated)
2. Aya Expanse 32B9432B125KCC-BY-NC— (estimated)
3. Qwen 3.6 27B93.628B256KApache-2.0— (estimated)
4. Qwen 3.5 14B93.314B125KApache 2.0— (estimated)
5. Qwen 3.7 35B-A3B92.135B256KApache-2.0— (estimated)
6. Gemma 4 31B9231B250KApache-2.0— (estimated)

Best pick for your memory budget

The strongest model overall is rarely the right answer — what matters is the strongest model that fits the memory you have. These picks are re-ranked per tier, so each one uses its budget rather than simply being small.

MemoryTypical hardwareRecommended models
8 GBRTX 4060, RTX 3070, base MacBook AirGLM-4.7 9B (96.1)
Qwen 3.5 7B (92.1)
GLM-6 9B (89.9)
12 GBRTX 3060 12 GB, RTX 5070GLM-4.7 9B (95.1)
Qwen 3.5 14B (94)
Qwen 3 14B (92.1)
16 GBRTX 5080, RTX 4080, RX 9070 XTQwen 3.5 14B (93.8)
GLM-4.7 9B (93.5)
Mistral Small 3.1 24B (92.6)
24 GBRTX 4090, RTX 3090, RX 7900 XTXAya Expanse 32B (96.9)
Qwen 3.6 27B (96.5)
Qwen 3.7 35B-A3B (95)
48 GBRTX 6000 Ada, MacBook Pro M4 Max 48 GBNemotron 70B Instruct (95.3)
Aya Expanse 32B (94.7)
GLM-5.1 72B (94.5)
128 GB+Mac Studio, DGX Spark, multi-GPUQwen 3.5 122B-A10B (MoE) (93.8)
GPT-oss 120B (93)
Qwen 3.5 122B-A10B (92.4)

How this ranking works

Split roughly evenly between reasoning and creative weight, since translation is simultaneously a fidelity and a fluency problem. Multilingual `bestFor` tagging carries more signal here than anywhere else, so the tag bonus frequently decides ranking between otherwise comparable models.

Worked example — GLM-4.7 9B: capability 85.4 × 0.414, quality 84.3 × 0.243, context 100 × 0.158, license 100 × 0.038, accessibility 100 × 0.146 + 6 tag bonus (multilingual, chinese).

Requirements applied: context floor 8,192 tokens (ideal 65,536), quality floor 50, licence weight 0.35, latency weight 0.6.

Running translation and multilingual work locally

FAQ

What is the best local LLM for translation and multilingual work?

GLM-4.7 9B, scoring 96.1/100 against this workload's published requirements. 111 models qualified.

What hardware do I need for translation and multilingual work?

A credible answer starts at 8 GB of memory. Larger budgets unlock materially stronger models — the table above lists the best pick at each tier.

How were these models ranked?

Split roughly evenly between reasoning and creative weight, since translation is simultaneously a fidelity and a fluency problem. Multilingual `bestFor` tagging carries more signal here than anywhere else, so the tag bonus frequently decides ranking between otherwise comparable models.

Hardware for This Workload

Related Workloads

Top Pick

Tools

← All workloads | Check your hardware