Orbix

The ORBIX Team thesis AI is sold at list price. The market sells it for less. Every major model has a list price set by the lab that made it. But a lot of AI capacity is already paid for: unused credits, committed spend, spare rate limits on keys that teams pay for every month. That capacity gets resold, and it trades far below list. Right now the AI resale market carries 10,230 seller listings across 99 of our 100 models, at a median 95% under list on output. In the last 24 hours it served 4.4M requests. The supply is real. What's been missing is a simple, safe way to use it. That's ORBIX. Why it matters for developers AI is becoming the biggest bill in a modern product, and agents make it bigger: they run all day and read far more tokens than any human would. You could buy from the resale market yourself. You'd have to find sellers, decide who to trust, handle the ones that fail, and juggle keys and billing for every model. With ORBIX you change one base URL: One key for 100 models: Claude, GPT, Gemini, DeepSeek, Qwen, GLM, Kimi, Grok and more. OpenAI and Anthropic compatible, with streaming and tool use. No code rewrite. Every request goes to the cheapest trusted seller at that moment. If no seller can take it, it falls back to OpenRouter, so production doesn't break. No subscription. Top up with ETH or USDG and pay per request. An overnight agent run, 10M tokens in and 2M out, cost $0.044 on GPT-5.6 Luna at today's market price. List price: $4.40. Why it matters for sellers Most teams pay for more AI capacity than they use. Unused credits expire, spare quota sits idle, and that's money gone. On ORBIX you list that capacity from the app, set your own price and daily limits per model, and get paid in USDC on Base, straight to your wallet, for every request you serve. Your provider key passes through to the marketplace and is never stored by ORBIX. ORBIX brings the demand. Every request goes to the cheapest trusted offer, so a sharp price wins volume. Why there is a token Market prices are already low. $ORBIX makes them lower for the people who back the network. Every $ORBIX trade pays a 1.2% creator fee into a public treasury. That fee funds a daily discount pool, and holders spend it as up to 80% off market prices, depending on how much they hold: 10% at 500K, 30% at 1M, 60% at 5M, 80% at 10M. No staking, no lockups. Hold it in your wallet and your key gets the discount. Usage gives people a reason to hold. Holding and trading fund the pool. A bigger pool makes AI cheaper for holders, which brings more usage. That's the loop. Why we're bullish on the product Demand compounds. Agents don't sleep. 4.4M requests a day already flow through the resale market, and most of them run on cheap, fast models at volume. Supply competes. 10,230 listings and growing. Models with 100+ sellers sit at a median 99% under list. More sellers means lower prices, and lower prices bring more buyers. Zero margin is a hard price to beat. We pass through exactly what the market charged. Nobody can undercut that by adding a markup. Discounts come from real fees, not emissions. The pool is funded by actual trading, in ETH, on-chain. Everything is checkable. Every response carries its list cost, market cost, discount and charged amount in the headers. The treasury is a public address. We ship fast. This week alone: 83 new models, USDG top-ups, live market prices across the site, a light theme, and a starter credit for holders. This is our view of the product and the market, not financial advice. Where we are today Live on Robinhood Chain 100 models, 99 with trusted sellers below list What we believe AI should cost what the market says, not what the list says. Every number should be checkable, so it's in the headers and on-chain. The people who back a network should be the ones who save the most from it. We're early, and we're building in public. Try it, check the receipts, and tell us what to build next. CA: 0x60633794df09856b339eea8bd741d749596b5230 useorbix.app

Private chat now has an uncensored model: Gemma 4 Uncensored. Most AI models are tuned to refuse anything that sounds sensitive. Ask about drug interactions, a lawsuit you're in or how a known exploit works, and you often get a lecture instead of an answer. An uncensored model drops those reflexes. It answers plainly where a regular model hedges or says no: - Health questions you'd rather not ask a person, answered directly - Legal and money situations, without "consult a professional" as the whole reply - Security research: how attacks work, so you can defend against them - Dark or adult themes in fiction, without the story being cut off - Blunt feedback on your writing, code or business plan And it runs in Private chat: end-to-end encrypted inside a sealed enclave, so nobody outside it can read the question, not even us. Nothing is saved. It's still bound by the law and our terms. Uncensored means fewer refusals, not anything goes. $0.19 / $0.88 per 1M tokens, holder discount applies. useorbix.app/app/private

We’re building ORBIX on Robinhood Chain to make AI cheaper to use. One API key for Claude, GPT, Gemini and more. Sellers compete on price, and $ORBIX trading fees fund extra discounts for holders. useorbix.app

New update: smarter ORBIX Agents Your agents now spend less and do more: - Checks before it thinks. Watchers look at prices, pages and wallets first and only call a model when something changed. A quiet check costs $0. - Auto fits your budget. It picks a model that fits your daily limit: $0.0011 a run under a $0.10 limit, down from $0.042. - Cached lookups. The same search within 10 minutes is free. - Long-term memory. Research agents keep what they found and recall it by meaning on the next run. - Self-check. Before answering, the agent checks the format, missing parts and links, and fixes what's off. - Longer tasks. A run can now take up to 10 minutes, in parts, and retries a step that fails. Live now: useorbix.app/app/agents

Meanwhile, our community asking when we're getting verified, where the blue check went, and when it's coming back for the 100th time

Reminder: free $0.50 to test ORBIX Every wallet can claim $0.50 of credit, no tokens needed. Spend it on any of our 200 models, the playground, agents or the new Private chat. Connect your wallet at useorbix.app/app and press Claim. One claim per wallet, for the first 100 wallets.

The Grind Never Stops.

What does "private" mean in Private chat? In a regular AI chat, your messages travel as plain text. The app can read them, the company hosting the model can read them, and so can anyone who breaks into their servers. Private chat works differently: 1. Your browser checks that the model runs inside a secure enclave: sealed hardware (Intel TDX + NVIDIA) that even its host can't look into. 2. It encrypts every message to that enclave. 3. ORBIX passes on ciphertext it can't read. 4. The answer comes back encrypted and opens only on your device. Nothing is stored, and the chat is gone when you close the tab. useorbix.app/app/private

New update: Private chat A new tab in ORBIX: chat with 12 confidential AI models that run inside secure enclaves (Intel TDX + NVIDIA). Your browser checks the enclave first, then encrypts every message to it. ORBIX only relays ciphertext, so nobody outside the enclave can read your chat, not even us. Nothing is saved: close the tab and it's gone. Models include Kimi K3, GLM 5.3, DeepSeek V4 Flash, GPT OSS 120B and an uncensored Gemma 4. You pay the list price per token, and your holder discount applies, up to 80% at the top tier. useorbix.app/app/private

One agent run, two bills. Our AI news digest searched the web 3 times and wrote 5 headlines. At list prices: $0.0096, and 61% of it was the model. On ORBIX: $0.0035. The model cost $0.000059 instead of $0.0059. Search stayed about the same. So the model is now the cheap part. Every run shows the cost of each step, so you can see where the money goes. useorbix.app/app/agents

We're currently catching up on all the DMs we've received here on X. We haven't been actively responding to DMs until now mainly for security reasons, but we're going through them and will do our best to answer everyone's questions. Feel free to reach out, just please avoid spamming or sending the same message multiple times. Appreciate your patience and support!

Agents, by the run. Five agents you can build on ORBIX today, each priced from one real test run: Coin alert, every 30 min: $0.0009 Release watcher, hourly: $0.0023 Morning crypto brief, daily: $0.0024 AI news digest, daily: $0.0035 Research desk, per question: $0.0073 Checking your coins 48 times a day comes to about $1.27 a month. useorbix.app/app/agents

Binance hacks, wars, markets dumping, the entire timeline on fire. Meanwhile, us building ORBIX: This is fine. useorbix.app

We’re working on some exciting updates for ORBIX, coming soon. Our goal is to make AI inference more accessible, affordable, and seamless for developers and AI agents. We’re also listening to the community. If you have ideas, feature requests, or anything you’d like to see on ORBIX, drop them below. We’re always open to feedback and ready to turn good ideas into real features. useorbix.app/

What can you build with ORBIX Agents? Here's one, start to finish. An AI news digest: every morning at 8:00 UTC it sends the 5 biggest AI stories of the last 24 hours, one line each, with a link. One message to the builder, and it filled in the card itself: model, web search, schedule, a $0.10 daily limit. The test run took 65 s and cost $0.0035 ($0.0096 at list). That's about 10 cents a month. Build yours: useorbix.app/app/agents

New update: ORBIX Agents Build your own AI agent right in ORBIX. Tell it the job in a chat, pick from 200 models or let it pick them itself, set a schedule and a daily limit, and call it from your own code. It bills to your key at ORBIX marketplace prices, and every run shows its cost next to list price. This is the first version. We'll keep building on agents from here. How to set up each part is in the article below. useorbix.app/app/agents

Huge thanks to the @X team for keeping our verification under review for 5 days straight. Really enjoying the suspense. Appreciate the dedication. 🙏

Next on ORBIX: build your own AI agent. Tell it what you need in a chat, and it sets itself up: the task, when to run, what data to use, which models and how much it may spend. - 10 presets to start from: routines, web research, X research, page and price watchers, on-chain watch, news digests, data extraction, content drafts, and an API endpoint for your own bots - One model, the cheapest one that fits, or a combination: a strong model plans and a cheap one does the work, or two answer and a third checks - Runs at market prices on the same key and balance you already use Coming soon. useorbix.app

We just passed 300 followers on X. Thank you for being here early, so here's something back. Free credit: $0.50 of AI usage, on us. No $ORBIX needed. Spend it on any of our 200 models: Claude, GPT, Gemini, DeepSeek and more. How to claim: 1. Open useorbix.app/claim 2. Connect your wallet 3. Press Claim $0.50 It lands in your balance right away. Try it in the Playground, or create a key and use it from your own code. One claim per wallet. The wallet needs at least one past transaction on Robinhood Chain, Ethereum, Base, Arbitrum or BNB Chain. The round closes when its budget runs out. useorbix.app/claim

Got API credit you won't use? Sell it on ORBIX. List spare capacity from a provider key you already pay for. 11 providers supported, including OpenRouter, Together AI, Fireworks, DeepSeek and Groq. 1. Paste the key and press Check key. You see every model it can serve. 2. Pick models, your discount off list (1 to 90%) and a daily cap. 3. Get paid in USDC on Base, straight to your wallet, once you've earned $5 or every 72 hours. ORBIX takes no cut. Your key goes to the marketplace encrypted, and ORBIX doesn't store it. The Sell page shows what you're owed, what's on its way and every payout with its Base transaction. Only list keys whose provider terms allow reselling, and set a spending limit on the key first. useorbix.app/app/sell