💻 Coding & Dev · Paid

BF16

Uncensored LLM API for coding agents and personal agents

1
Views
0
Saves

About BF16

BF16 is an uncensored LLM API for coding and personal agents, serving OpenAI- and Anthropic-compatible endpoints with 1M-context DeepSeek, GLM and Kimi models whose refusal direction has been projected out of the weights. Pay-as-you-go from a prepaid dollar balance with $0.50 free credit and no card required, and prompts are never stored or logged.

Updated 11/17/2026

Key Features

  • OpenAI- and Anthropic-compatible endpoints — drop-in for Claude Code, Codex, OpenCode, Goose and Zed via ANTHROPIC_BASE_URL
  • 1M-token context on every served model
  • Uncensored DeepSeek V4.1 Flash, GLM-5.3, GLM-5.3 Flash Vision and Kimi K3.1 Pro
  • Refusal direction projected out of the open-weight checkpoint — a weight edit, not a prompt jailbreak
  • Priced below official vendor list rates: DeepSeek V4.1 Flash from $0.26 per 1M input tokens and $1.00 per 1M output
  • Billing metadata only — no prompt, code or chat history written to disk, logged or used for training
  • Prepaid wallet with $10-$2,000 top-ups via Waffo and Dodo Payments; purchased credit never expires
  • Rate limits rise with total top-ups, from 3 concurrent requests and 180 RPM free to 100 concurrent and 6,000 RPM
  • Dedicated single-tenant H200 / B200 inference pods with custom LoRA adapters

Pricing Structure

  • Free trial — $0.50 credit — Automatic on verified signup, no card, 30-day validity
  • Pay-as-you-go — from $0.26 / 1M tokens — Prepaid wallet, $10-$2,000 top-ups, never expires
  • Dedicated GPU pods — Custom — Single-tenant H200/B200, SLA, custom LoRA

See current plans at BF16.

Why People Use It

Long agent runs die when a safety classifier trips mid-session: the thread is poisoned, the context is gone and hours of work restart from zero. BF16 hosts abliterated open-weight models behind the endpoints your tools already speak, so auditing your own codebase or locking down your own server finishes without a vendor lecture. Pricing is per token and below the vendors' own rates, and prompts are not stored — which is what makes it usable on proprietary code.

See BF16 in Action

BF16 screenshot

Reviews

⭐

No reviews yet. Be the first!