TopoLabs
[Changelog]

What's new in Topo.

Releases across the API, the router, settlement and the node client, newest first. Breaking changes ship as a new dated API version.

  1. SiteLatest

    Topo in Chinese

    The site and the docs now ship a complete Simplified Chinese edition. Switch with the EN / 中文 toggle in the corner; your choice is remembered on this device.

    • Every page and every docs article, in both languages.
    • Code samples, model IDs and API fields stay as they are.
    Open the docs
  2. Network

    Topo Network mainnet

    The network leaves beta. Every request is routed to a staked node, verified by independent attestors and settled on-chain with a receipt. Nothing changes in your code.

    • 120+ models behind one OpenAI-compatible API.
    • Staked GPU operators in 40+ regions.
    • Per-token settlement in USDT, ETH or SOL on Base and Solana.
    • API version 2026-10-01 is the new default for new keys.
    Read the announcement
  3. Router

    Hedged routing, p50 182 ms

    The router now keeps a runner-up node in reserve and starts it when the first token is late. You are billed once, for the stream you receive.

    • Network-wide median time to first token down from 240 ms to 182 ms.
    • Scoring uses live queue depth instead of a five second average.
    • New x-topo-route response header with the serving node and timings.
    Read the deep dive
  4. Verification

    Sampled re-execution and signed receipts

    A sample of requests is re-run by independent attestors on different hardware, and every response now carries a receipt signed by the serving node.

    • Three of three attestors must agree before a sampled response settles.
    • Higher sampling rates for new nodes and nodes with recent misses.
    • Receipts in the response body, the dashboard preview and on-chain.
    How verification works
  5. API

    Streaming with usage and receipts

    Set stream: true to receive server-sent events in the OpenAI format. The stream closes with a usage chunk and the receipt for the request.

    • Works unchanged with the OpenAI SDKs for Python and TypeScript.
    • Close the connection to cancel and pay only for tokens already generated.
    Streaming docs
  6. Payments

    Pay in TOPO, save 20%

    Requests paid in TOPO now cost 20% less, converted at the 15-minute time-weighted price from the network oracle. The discount stacks with plan discounts.

    • The rate used is written on every receipt.
    • Per-token settlement in USDT is unchanged and remains the default.
    TOPO discount
  7. Node

    Run a node on a single 4090

    The node client now supports consumer GPUs. A 24 GB RTX 4090 can serve quantized open models up to about 32B parameters, plus embedding and image models.

    • One container image with a 10 minute benchmark on first start.
    • Minimum stake of 2,500 TOPO per GPU, paid out daily in USDT.
    Run a node
  8. Docs

    Docs, quickstart and API reference

    Topo documentation is live: a five minute quickstart, the full API reference, SDK guides and the node operator handbook, all searchable.

    • Code samples in cURL, Python and TypeScript on every endpoint.
    • Error codes, rate limits and settlement explained end to end.
    Open the docs
Release notes by email, about twice a month.support@topolabs.ai