Deepseek 🐋
Qwen3.8 27B: Same Model, Three Harnesses, One Clear Winner - 1. Deepseek, 2. Pi, 3. Hermes - github
# API Pricing ⮺
# Off-peak ⮺
Your day, priced
Checking...
--:--
FRANCE · EUROPE/PARIS
NOW
00:00
06:00
12:00
18:00
24:00
Off-peak
Peak
OFF-PEAK
$0.15
Input / 1M tokens
Cached input
$0.003
Output
$0.60
PEAK
$0.30
Input / 1M tokens
Cached input
$0.006
Output
$1.20
Prices in USD per million tokens ·
DeepSeek official pricing
| Usage | DeepSeek Direct (Off-peak) | DeepSeek Direct (Peak) | ClinePass Reference |
|---|---|---|---|
| Input (per 1M tokens) | $0.15 | $0.30 | $0.30 |
| Cached input (per 1M tokens) | $0.003 | $0.006 | $0.006 |
| Output (per 1M tokens) | $0.60 | $1.20 | $1.20 |
Prices in USD per 1 million tokens.
# Harness
It is built on an everything-is-a-plugin architecture and powered by Cordis, whose design is described in A Programming Paradigm for Spatiotemporal Composability.
# VSCode
# Setup
$ node --expose-internals \
/home/yves/.npm/_npx/1e7f6d9597241db0/node_modules/@deepseek-ai/dsh/lib/bin.js \
web --no-open # --no-open prevent opening the page in browser
# LM Studio
To use LM Studio with this harness you have to configure a custom adapter.
need to put a password in the API key It won’t be used, but it has to be filled so the harness work.
Provider ID: lmstudio Display name: LM Studio Base URL: http://yves-lab:1234/v1 API protocol: openai-completions API key: generate a password
# v4.1 Flash (LLM)
- DeepSeek choque encore toute l’industrie de l’IA (actu IA)
- KV Cache sur SSD
# Running Local
# 12GB GPU
- Run FULL DeepSeek v4.1 Flash on a 12GB GPU - at 7 token / s.
# see also
Written on September 11, 2026, Last update on
agentic-AI
harness