ai.canshu.com
中文
VRAM Calculator Advisor & TCO API Pricing Compare Models Leaderboard GPUs

Search models, GPUs, and tools

↑↓ Navigate · Enter Open · ★ Watchlist prioritized ai.canshu.com
Synced daily · 270 LLMs · 31 GPUs · 375 Evals

Size Your LLM Hardware & API Bill

Exact KV & LoRA memory sizing from HuggingFace config.json, real-workload API billing, and 12-month GPU break-even curves.

Popular:
Live VRAM GaugeQ4_K_M · FP16 KV
Qwen3-14B × RTX 4070 Ti Super
12.29/ 15.7 GB usable
Smooth Fit
W 8.97G + KV 2.68G+3.41 GB free
Context:
~46.9 tok/s
Max runnable ctx: 36KOpen Full Calculator

Price & Model Release Radar

11-17New · $2/$10
OpenAI Launches GPT-6.2 Astra & GPT-6.2 Sol

Sol priced at $2/$10 per 1M tokens for daily coding; Astra flagship at $8/$40

11-16-50% Off-Peak
DeepSeek-V4.1-Flash & V4-Pro 50% Off-Peak Tier Active

Off-peak window cuts V4.1-Flash to $0.15/M input ($0.015/M cached) and $0.60/M output

11-1524GB Sweet Spot
Qwen3.8-27B Open Weights + Bailian Omni-Flash Rate

Fits 24GB GPUs cleanly at Q4_K_M; Bailian Omni-Flash API at ¥0.76 / ¥2.60 per 1M

11-12$1.5 / $7.5
Gemini 4 Hydrogen 1M Context Tier at $1.5 / $7.5

1M multimodal context window with $1.50/M input and $7.50/M output

11-10FREE Tier
ERNIE-5.5-Flash & Ling 3.2 Flash Free Developer Tier

Zero-cost developer API tier with CN direct access and tool calling support

11-09-90% Cache
Claude Fable 5.2 Leads Coding Eval · 90% Cache Discount

Repeated repository context billed at 10% of base input rate under Prompt Caching

Decision Tools

Popular GPUs & Devices

All 31 →