FITMYLLM · NOVEMBER 4, 2026 · OPEN METHODOLOGY · COMMUNITY BENCHMARKS
LIVE · UPDATED DAILY
604.34B
Parameters (27B active)
Quantization Options Select your GPU for speed estimates Context length: 4K 8K 16K 32K 64K 128K 256K 512K 1M
Quant Bits VRAM @ 16K Quality IQ2_XXS 2.38 181.2 GB
180.3 + 0.9 KV
low IQ2_M 2.93 222.7 GB
221.8 + 0.9 KV
low Q2_K 3.16 240.1 GB
239.2 + 0.9 KV
low IQ3_XXS 3.25 246.9 GB
246.0 + 0.9 KV
low IQ3_XS 3.5 265.8 GB
264.9 + 0.9 KV
low Q3_K_S 3.64 276.4 GB
275.5 + 0.9 KV
low IQ3_M 3.76 285.4 GB
284.5 + 0.9 KV
low Q3_K_M 4 303.6 GB
302.7 + 0.9 KV
low Q3_K_L 4.3 326.2 GB
325.3 + 0.9 KV
moderate IQ4_XS 4.46 338.3 GB
337.4 + 0.9 KV
moderate Q4_K_S 4.67 354.2 GB
353.3 + 0.9 KV
moderate Q4_K_M 4.89 370.8 GB
369.9 + 0.9 KV
good Q5_K_S 5.57 422.2 GB
421.3 + 0.9 KV
good Q5_K_M 5.7 432.0 GB
431.1 + 0.9 KV
good Q6_K 6.56 497.0 GB
496.0 + 0.9 KV
excellent Q8_0 8.5 643.5 GB
642.6 + 0.9 KV
lossless FP16 16 1210.1 GB
1209.2 + 0.9 KV
lossless
Select your GPU above to see speed estimates and compatibility for each quantization.
Too big for a single GPU — plan a multi-GPU deployment
Even the lightest quant needs ~181 GB. Size GPUs, replicas, TCO and scaling for a production setup. Open in Enterprise →
▸ READY TO RUN THIS? RENT BY THE HOUR
RENT A GPU AND RUN STEP 5 PREVIEW 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 — 180.3 GB VRAM IQ2_M — 221.8 GB VRAM Q2_K — 239.2 GB VRAM IQ3_XXS — 246.0 GB VRAM IQ3_XS — 264.9 GB VRAM Q3_K_S — 275.5 GB VRAM IQ3_M — 284.5 GB VRAM Q3_K_M — 302.7 GB VRAM Q3_K_L — 325.3 GB VRAM IQ4_XS — 337.4 GB VRAM Q4_K_S — 353.3 GB VRAM Q4_K_M — 369.9 GB VRAM Q5_K_S — 421.3 GB VRAM Q5_K_M — 431.1 GB VRAM Q6_K — 496.0 GB VRAM Q8_0 — 642.6 GB VRAM FP16 — 1209.2 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:604b-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 Step 5 Preview
Build Hardware for Step 5 Preview ▸ COLOPHON FITMYLLM · INDEPENDENT · DATA-DRIVEN
FITMYLLM · EST. 2025 · © 2026
RECOMMENDATIONS FROM PUBLISHED MATH, CORRECTED BY THE COMMUNITY — 30.
▸ SPEC SHEET
Step 5 Preview — 604.34B MoE. ▸ SPECIFICATIONS
PARAMETERS 604.34B (27B active)
ARCHITECTURE Mixture of Experts
CONTEXT LENGTH 1024K tokens
CAPABILITIES chat, coding, reasoning, multilingual, vision, math, agentic, tool_use
RELEASE DATE 2026-10-15
PROVIDER StepFun
FAMILY other ▸ VRAM REQUIREMENTS
QUANT BPW VRAM QUALITY IQ2_XXS 2.38 180.3 GB 65% IQ2_M 2.93 221.8 GB 75% Q2_K 3.16 239.2 GB 78% IQ3_XXS 3.25 246.0 GB 82% IQ3_XS 3.5 264.9 GB 84% Q3_K_S 3.64 275.5 GB 85% IQ3_M 3.76 284.5 GB 86% Q3_K_M 4 302.7 GB 88% Q3_K_L 4.3 325.3 GB 90% IQ4_XS 4.46 337.4 GB 92% Q4_K_S 4.67 353.3 GB 93% Q4_K_M 4.89 369.9 GB 94% Q5_K_S 5.57 421.3 GB 96% Q5_K_M 5.7 431.1 GB 96% Q6_K 6.56 496.0 GB 97% Q8_0 8.5 642.6 GB 100% FP16 16 1209.2 GB 100%
§ 01 BENCHMARK SCORES
HLE 46.5
aa_scicode 58.9
aa_lcr 88.3
aa_terminal_bench_v4 33.3
§ 03 COMPATIBLE GPUs
2 @ Q4_K_M Feedback