Claude Mythos has arrived (and is called Claude Fable).
- Blows every other model out of the water, especially in terms of programming.
- Heavily guardrailed because it's that good at cyber & bio.
- $10/$50 per mln tokens input and output.
Go try it out, live on NanoGPT!
@EmotionalThrif2@vectara We just tested GPT Image 2 multi-image editing and it worked on our side. No known broad issue right now. Could you send your support key/request ID and the model + prompt you used? Also note some edit models only support one input image, but GPT Image 2 should accept multiple.
Model benchmark pages now include @vectara's hallucination benchmark data where available.
Use the benchmark source selector to compare models on hallucination-focused evaluation alongside the existing Artificial Analysis and LMArena views.
x402 API payments now support @solana USDC!
Developers can choose `x402-solana-usdc` from a 402 quote, sign the Solana exact payment, and replay the original request with `X-PAYMENT`.
Works for all chat models, all image models, and much much more!
Gemini chat models now support native audio input. Upload or send audio in a chat/API request and supported Gemini models can reason over the spoken content directly, instead of needing a separate transcription step first.
⚡This Saturday on the THORChain Podcast | Ecosystem Spotlight.
@NanoGPTcom joins @KentonC137 and @patriotsounds live at 10AM EDT / 2PM UTC.
NanoGPT provides private, pay-per-prompt access to ChatGPT and top AI models using Nano, crypto, or cards.
Jump on stage and ask your questions live: https://t.co/AUD5BPli7X
NanoGPT Help is now available as a dedicated chat model for questions about NanoGPT.
Ask it about billing, subscriptions, models, API usage, privacy, payments, or other site features, and it will answer with relevant help sources.
Note this is still kind of in beta, so let us know if anything seems off.
⚡ Saturday 13th June, on the THORChain Podcast | Ecosystem Spotlight at 10AM EDT / 2PM UTC.
@KentonC137 and @patriotsounds are sitting down with @NanoGPTcom .
NanoGPT provides anonymous, private, pay-per-prompt access to ChatGPT and other leading AI text and image models, with payments supported via Nano, cryptocurrencies, and credit cards.
Jump on stage and ask your questions live: https://t.co/AUD5BPli7X
Linkup Research Low, Medium, High, and XHigh are now available!
These models run source-grounded web research for complex questions that need citations, multi-source synthesis, or deeper investigation.
They can take several minutes to return, so they're best for research tasks rather than quick chat.
API calls can now opt into paid inline moderation.
Add moderation: true to text, image, or video generation requests to have NanoGPT run an input safety preflight.
You can also select a specific moderation model with moderation: <model-id> or moderation-model: <model-id>.
If moderation flags the input, we charge only the moderation check, skip generation, and return content_policy_violation; if it passes, generation continues normally.
Solana deposits now also support USDT.
You can choose USDT on Solana from the deposit flow, get a Solana payment link for the selected asset, and fund your NanoGPT balance through the same Solana deposit option.
NVIDIA Nemotron ASR Multilingual is now available for speech-to-text. It is a low-latency transcription model with automatic language detection, native punctuation, capitalization, and support for 40 language-locale combinations.
Model pages now show average output tokens over the last 7 days. You can also see how each model compares against others by percentile to spot verbose models versus efficient ones.
Global Memory is getting smarter while staying optional and opt-in.
Saved memories can now track freshness, so temporary facts like trips, current projects, or short-term plans can stop affecting chats after they expire.
Memory is also more selective in larger memory sets: small memories are still recalled fully, while larger memories are ranked against the current request so unrelated details are less likely to clutter the prompt.
Important context such as user instructions, response-style preferences, allergies, dietary needs, medical/accessibility constraints, and family constraints is protected so it remains available when it matters.
The Memory settings panel now shows freshness counts and includes cleanup for expired memories.
The result should be better personalization, fewer stale assumptions, and lower context overhead if you choose to use Global Memory.
@rinse_12 It is on by default now if you pick a TEE model that supports it - the ones with a shield and private next to the name are the ones that support it!
Private mode is now available in NanoGPT!
TEE mode, fully verifiably private.
Select a supported TEE model, enable Private Mode in the model picker, and your browser encrypts the chat request before NanoGPT receives it.
The NanoGPT Batch API now supports high-volume asynchronous /v1/chat/completions jobs for supported GPT, Claude, and Gemini chat models, at 50% discount.
Use via the API or via the Batch API page on our website!
Zero-fee Solana top-ups, 370% faster and lower minimums!
We just launched a major upgrade to our Solana payments that slashes every friction.
This was the first step of us in-housing most of the crypto we accept, to give you more privacy and minimum fees.
$SOL, $USDC and $USDT are natively accepted
Which network should be next?
Developers can use Private Mode from OpenAI-compatible API clients through the local proxy package: NANOGPT_API_KEY=sk-your-key npx @nanogpt/private-mode, then point clients at http://127.0.0.1:8787/v1.