Log. Releases, research, specs and models, dated and newest first.
Everything the lab puts out, with the day it went out. Research and protocol drafts are also collected under Research.
September 2026
- , Site: Twigpine: a new name, the same network and keysThe rename changes what we are called, not what you signed.
- , Research: How Darwin decides which model to keepThe acceptance rule since generation 10: held-out accuracy, Brier score and log loss, pre-registered challengers, 95% Wilson intervals.
- , Spec: GLIP-01, 02, 09 and 10Core protocol, node information document, visibility and withheld content, delegated capabilities. Twigpine protocol drafts, public domain.
- , Release: Zero 0.9.0Resume sessions, attach to a running terminal, and reliability work. (external site)
- , Release: Darwin era 2Settlement, accounting and validation rebuilt.
August 2026
- , Release: Console early accessClaim your handle. (external site)
- , Release: Pay per request over x402On Base mainnet, with no API key.
- , Release: Explorer answers agents in markdownllms.txt and skill.md published.
- , Release: Playground: bring your own Grok keyPlayground builds can run on your own Grok key; Grok 4.6 is the default. (external site)
- , Release: Anthropic Messages API on the gatewayA /v1/messages endpoint alongside /v1/chat/completions.
- , Release: Credits, cost headers and a live catalogA non-streaming response carries its cost in a header; the catalog is served on /v1/models.
- , Release: Free models retired; Nemotron 3 Ultra stays freeEvery catalog model is paid except Nemotron 3 Ultra, which keeps a rate-limited free tier.
- , Model: Nemotron 3 Ultra added to the gatewayFrontier reasoning MoE · 1M context.
- , Model: Macaron V1 Venti added to the gatewayFlagship routed MoL on GLM-5.2 · 1M context.
- , Release: Network node 0.7.1Replication hardening: repository paths are checked before a peer can write them. (external site)
July 2026
- , Model: Macaron V1 Tall added to the gatewayRouted MoL personal-intelligence model · 262k context.
- , Model: Ling 3.0 Flash added to the gatewayToken-efficient agentic MoE · 262k context.
- , Model: Kimi K3 added to the gatewayFrontier multimodal reasoning · 1M context.
- , Model: Tencent HY3 added to the gatewayMultimodal model · 262k context.
June 2026
May 2026
Showing 25 of 25 entries · RSS