This model is based on DeepSeek-V4.1-Flash-1017-lossless-CSF by Local Inference Lab, Inc., a non-profit organization, available at https://huggingface.co/local-inference-lab/DeepSeek-V4.1-Flash-1017-lossless-CSF. DeepSeek-V4.1-Flash-1017-lossless-CSF is licensed under the Local Inference Lab License, Version 1.0.DeepSeek V4.1 Flash 1017 Abliterated CSF (jcartu) is now available. jcartu’s CSF derivative of DeepSeek V4.1 Flash 1017 applies norm-preserving abliteration to 132 dense tensors. It accepts text, supports tool calling and structured JSON output, and offers none, low, high and max reasoning effort. Input and output share a 1,048,576-token context window, with up to 943,718 output tokens. This offering uses jcartu/DeepSeek-V4.1-Flash-1017-abliterated-lossless-CSF.Input costs $0.315, output $1.05 and cached input $0.021 per million tokens.https://nano-gpt.com/r/qzFYzwRg
- 3691 FOLLOWERS
- 205 FOLLOWING
- MEMBER OF 30 CLUBS
Mercury Decide by Inception is now available on NanoGPT through @club/v1/decisions. Ask named questions over shared context and receive binary probabilities, categorical choices, ordinal scores or exact text spans with confidence. Supports text, JSON and up to four inline PNG images; extraction is text-only and supports up to 26 fields. Input costs $0.02 per million tokens, with no output charge.
How I verify regional risk before an alert reaches me.
👋FREE👋Clover 2 is Kitani's new 27.8B model: a dense base with seven LoRA experts and trained per-token routing between them. It handles text chat, with optional reasoning in Clover 2 Thinking. It has a 32,768-token context window and up to 8,192 output tokens; tool calling is not available.Both are free on NanoGPT until November 24, 2026 03:33 UTC, with a per-account request cap during the free week.
Qwen 4 Max is now available for coding, knowledge work, data analysis, and long-running agent workflows. It accepts text, image, and video input and returns text with selectable reasoning, tool calling, and structured output. Its context window is approximately one million tokens; non-thinking responses support up to 131,072 output tokens, while thinking and final output share a 262,144-token budget. Choose none, low, high, or extra-high reasoning effort, with high as the default for the thinking variant.
How I am tracking regional risk today, without drowning in news alerts.1. Verify the news first. Reuters, AP, official sources only. State media claims stay unconfirmed until corroborated.
2. Feed only the verified facts to a small decision model, eu-jeff-latest on NanoGPT. It scores direct risk 0-4, plus a yes/no: is this significant now? Jeff is available on NanoGPT:
3. Ping only on a material change, or a score of 2+ with verified facts behind it. Everything else stays silent and gets logged.This morning: alerts in a neighbouring state plus intercepts nearby, no target where I am. Jeff scored it 1.02/4, stable, 1.5% significant. Elevated, not a warning.The model is a helper, not the decision. The facts do the work. Each score costs about $0.00003.#Nostr #AI #Risk #DecisionModels
Turn on Embed prompt in downloaded images in Settings → Privacy & analytics, and images you download from NanoGPT carry their prompt, negative prompt, model, size and seed in the file. It uses the Automatic1111 format, so SD Prompt Reader and Civitai can read it. It works for PNG, JPEG and WebP from the download button, your gallery, gallery ZIP exports and images made in chat, and never changes the pixels. It is off by default for privacy, because anyone you share a file with can read its prompt. Only the image prompt is written, never your chat messages.
When PII redaction is on for a zero data retention request, NanoGPT now always redacts it with Tinfoil, even if Grepture was chosen in the request, the API key or your settings, so the request never passes through Grepture. Detection runs in Tinfoil's attested, encrypted enclave. A request requires zero data retention when its API key is set to it (or the key has no setting and your account is), when it is billed to a team set to zero data retention, or when it sends provider.zdr: true. Zero data retention doesn't turn redaction on by itself. Tinfoil's redaction pricing applies.https://nano-gpt.com/r/qzFYzwRg
Uh I thought monad was a total failure on Farcaster, did not have lens x monad on my bingo card. In my view monad was dead lobg time ago 🤣
Jev 1.14 now accepts up to 1,000 inline PNG or JPEG images per Decisions API request, up from four, with up to 3.75 MiB of image data in a request of up to 4 MiB. Each image can be 280 to 1568 pixels per side. A 280-pixel image counts as about 240 input tokens and a 1568-pixel image as about 6,300, and an image request can use up to 262,144 tokens. Images are billed as input tokens at the usual Jev 1.14 rate, and zero data retention works the same way. Requests with moderation remain limited to four images.