deepseek-v4.1-flash-heretic

Uncensored DeepSeek V4.1 Flash 1017 (abliterated). 1,048,576-token context, image input, OpenAI-compatible.

1M contextvision no signupno KYC

Price

Input	Output	Notes
$0.4125 / M	$1.65 / M	flat across the whole context, no long-context surcharge

launch price — 50% off — half the usual rate while the service fills up; reverts to $0.825 / $3.30 per M

Problems, refunds or a custom rate: agent_smith@maxxspace.com — a human reads it.

Pay

Three ways in — pick one.

Pay with Telegram Stars

Stars carry a +35% price offset. Telegram and the mobile app stores keep roughly a third of a star sale, so $5 of credit costs 520 stars (≈ $6.76 at Telegram's seller rate) — you still receive the full $5 of credit. Buying stars inside the iOS/Android app costs more than buying them on Fragment, so the sticker price you see depends on your platform. Card and USDC carry no markup.

Prepaid with crypto

Rail	Address
USDC (Base)	`0x5Ab610B2C88438548b03511b943BeBF4C324348e`
BTC	`bc1qvrjtc9xuwnkq32aac493x627m9sgs65akvgsku`
XMR	`49uzZZdmksm3HB4XZbR4SpZ4FTcjbFf8zjYXE4NHLVNJTAVNYn7Yw4nZq4MTWQgUJCRRNqKAihf8eBd3f6abevSB4LK7svW`

Pay the exact amount, to the last decimal.
Invoices carry a small unique tag (e.g. 5.004317 USDC) so your transfer can be matched to your invoice automatically. Send that exact figure, in one transfer. A rounded amount (5.00), a different amount, or two separate transfers will not be matched, and no key is issued automatically — you would have to email the operator with your tx hash.

Minimum $1. Create an invoice first: you get the exact amount to send and a secret claim token. Then claim with the token (USDC ~1 min, BTC after 2 confirmations). Keep the token — anyone who has it can take the key. XMR: contact the operator.

Per request, no account (x402)

curl -s -X POST https://api.flashctx.xyz/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model":"deepseek-v4.1-flash-heretic","messages":[{"role":"user","content":"hi"}],"max_tokens":128,"stream":true}'
# 402 -> pay the quoted USDC amount -> retry with X-PAYMENT: <base64 payload>

Measured performance

Workload	Result (this node, 2026-11-11)
Prefill, single stream	4.5k tok/s at a 7K prompt · 8.4k at 28K · 10.8k at 85K
Time to first token	0.24 s short prompt · 1.6 s at 7K · 3.4 s at 28K · 7.9 s at 85K
Decode, one stream	~37 tok/s (300–400 token answers)
Decode, aggregate	~112 tok/s at 4 concurrent · ~188 tok/s at 8 concurrent

Plain streaming chat completions, no speculative decoding, measured with the public model on one 4-GPU node. Long prompts push prefill up; busy seats split the aggregate. These are the numbers to plan against, not peak marketing figures.

Use it

from openai import OpenAI
c = OpenAI(base_url="https://api.flashctx.xyz/v1", api_key="sk-…")
for chunk in c.chat.completions.create(model="deepseek-v4.1-flash-heretic", stream=True,
        messages=[{"role":"user","content":"hello"}]):
    print(chunk.choices[0].delta.content or "", end="")

Long answers: streaming recommended.
Output runs at roughly 37 tokens/s, so a full 8K answer takes a few minutes. With "stream": true you see tokens as they arrive. Non-streaming requests also work at any length: if the answer takes more than ~20 s the server sends the 200 headers early and pads the body with leading spaces until the JSON is ready (valid JSON; every standard client parses it). Set your client's read timeout to at least 10 minutes. Errors on such slow requests arrive as an {"error": …} body with status 200.

Images must be inlined as data: URIs. 8K output cap, 256K input cap, 6 concurrent seats. Prompts retained 14 days, never used for training.