FITMYLLM · AUGUST 17, 2026 · OPEN METHODOLOGY · COMMUNITY BENCHMARKS
LIVE · UPDATED DAILY
Quantization Options Select your GPU for speed estimates Context length: 4K 8K 16K 32K
Quant Bits VRAM @ 16K Quality IQ2_XXS 2.38 32.1 GB
31.7 + 0.4 KV
low IQ2_M 2.93 39.4 GB
38.9 + 0.4 KV
low Q2_K 3.16 42.4 GB
42.0 + 0.4 KV
low IQ3_XXS 3.25 43.6 GB
43.1 + 0.4 KV
low IQ3_XS 3.5 46.8 GB
46.4 + 0.4 KV
low Q3_K_S 3.64 48.7 GB
48.3 + 0.4 KV
low IQ3_M 3.76 50.3 GB
49.8 + 0.4 KV
low Q3_K_M 4 53.4 GB
53.0 + 0.4 KV
low Q3_K_L 4.3 57.3 GB
56.9 + 0.4 KV
moderate IQ4_XS 4.46 59.4 GB
59.0 + 0.4 KV
moderate Q4_K_S 4.67 62.2 GB
61.8 + 0.4 KV
moderate Q4_K_M 4.89 65.1 GB
64.7 + 0.4 KV
good Q5_K_S 5.57 74.0 GB
73.6 + 0.4 KV
good Q5_K_M 5.7 75.7 GB
75.3 + 0.4 KV
good Q6_K 6.56 87.0 GB
86.6 + 0.4 KV
excellent Q8_0 8.5 112.5 GB
112.1 + 0.4 KV
lossless FP16 16 210.9 GB
210.5 + 0.4 KV
lossless
Select your GPU above to see speed estimates and compatibility for each quantization.
Deploying for a team or in production? Size GPUs, cost & scaling in Enterprise → ▸ READY TO RUN THIS? RENT BY THE HOUR
RENT A GPU AND RUN SARVAM 105B NOW
Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.
Community Ratings Chat Coding Reasoning Creative Vision Roleplay Agentic
Loading ratings...
Run this model IQ2_XXS — 31.7 GB VRAM IQ2_M — 38.9 GB VRAM Q2_K — 42.0 GB VRAM IQ3_XXS — 43.1 GB VRAM IQ3_XS — 46.4 GB VRAM Q3_K_S — 48.3 GB VRAM IQ3_M — 49.8 GB VRAM Q3_K_M — 53.0 GB VRAM Q3_K_L — 56.9 GB VRAM IQ4_XS — 59.0 GB VRAM Q4_K_S — 61.8 GB VRAM Q4_K_M — 64.7 GB VRAM Q5_K_S — 73.6 GB VRAM Q5_K_M — 75.3 GB VRAM Q6_K — 86.6 GB VRAM Q8_0 — 112.1 GB VRAM FP16 — 210.5 GB VRAM
Ollama llama.cpp vLLM LM Studio KoboldCpp Jan Docker
▸ Easiest way to get started · Beginners
DOCS ↗ curl -fsSL https://ollama.com/install.sh | shCOPY
$ ollama run other:105b-q4_K_MCOPY
Tag may need adjustment — check ollama.com/library/other for available tags.
▸ SETUP GUIDE >_
Auto-setup with fitmyllm CLI Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.
Auto-detect GPU Live tok/s in chat Speed benchmarks 9 inference engines
GPUs that can run this model At Q4_K_M quantization. Sorted by minimum VRAM.
Find the best GPU for Sarvam 105B
Build Hardware for Sarvam 105B ▸ COLOPHON FITMYLLM · INDEPENDENT · DATA-DRIVEN
FITMYLLM · EST. 2025 · © 2026
RECOMMENDATIONS FROM PUBLISHED MATH, CORRECTED BY THE COMMUNITY — 30.
▸ SPEC SHEET
Sarvam 105B — 105B Dense. ▸ SPECIFICATIONS
PARAMETERS 105B
ARCHITECTURE Dense Transformer
CONTEXT LENGTH 32K tokens
CAPABILITIES chat, multilingual, reasoning
RELEASE DATE 2026-03-06
PROVIDER Sarvam AI
FAMILY other ▸ VRAM REQUIREMENTS
QUANT BPW VRAM QUALITY IQ2_XXS 2.38 31.7 GB 65% IQ2_M 2.93 38.9 GB 75% Q2_K 3.16 42.0 GB 78% IQ3_XXS 3.25 43.1 GB 82% IQ3_XS 3.5 46.4 GB 84% Q3_K_S 3.64 48.3 GB 85% IQ3_M 3.76 49.8 GB 86% Q3_K_M 4 53.0 GB 88% Q3_K_L 4.3 56.9 GB 90% IQ4_XS 4.46 59.0 GB 92% Q4_K_S 4.67 61.8 GB 93% Q4_K_M 4.89 64.7 GB 94% Q5_K_S 5.57 73.6 GB 96% Q5_K_M 5.7 75.3 GB 96% Q6_K 6.56 86.6 GB 97% Q8_0 8.5 112.1 GB 100% FP16 16 210.5 GB 100%
§ 01 BENCHMARK SCORES
MMLU-PRO 81.7
MATH 98.6
IFEval 84.8
SWE-bench 45.0
AIME 88.3
GPQA Diamond 78.7
HLE 10.1
AA Intelligence 18.2
AA Coding 9.8
aa_ifbench 34.4
aa_terminal_bench 1.5
aa_tau2 46.8
aa_scicode 26.4
§ 03 COMPATIBLE GPUs
30 @ Q4_K_M Feedback 