inclusionAI Ring-2.6-1T lands free on OpenRouter

// 2h agoMODEL RELEASE

inclusionAI Ring-2.6-1T lands free on OpenRouter

Ring-2.6-1T is a 1T-parameter-scale thinking model from inclusionAI now listed on OpenRouter with free access. The model is positioned for production agent workflows, with adjustable reasoning effort, 63B active parameters, and an emphasis on coding, tool use, and long-horizon task execution. OpenRouter’s listing frames it as a strong fit for agentic systems where latency, cost, and execution quality all matter.

// ANALYSIS

Hot take: this looks like a practical agent-model release more than a hype drop. The combination of free access, adjustable thinking effort, and explicit optimization for tool-heavy workflows is what makes it interesting.

–OpenRouter positions it as a 1T-scale model with 63B active parameters, which is a strong signal for capability without full compute overhead.
–The adaptive reasoning knob in high and xhigh modes is the most relevant feature for real agent builders.
–The model is being pitched for coding agents, complex reasoning pipelines, and long-horizon autonomous tasks, not just chat demos.
–The free OpenRouter window through May 15 lowers friction for immediate evaluation and side-by-side testing.
–Benchmarks cited by OpenRouter suggest it is competitive in agentic and search-heavy settings, which is where adoption will likely come from.

// TAGS

openrouterinclusionaireasoning-modelagentscodingtool-usefree-tier

DISCOVERED

2h ago

2026-05-11

PUBLISHED

2h ago

2026-05-11

RELEVANCE

9/ 10

AUTHOR

OpenRouter

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

NEWS45m ago

Grok Skills preview hints at task automation

xAI appears to be building a custom Skills system for Grok that lets users save reusable instruction sets and import them as `.zip`, `.skill`, or `.md` packages. The feature is still behind a flag, so this is a preview rather than a public rollout.

UPDATE1h ago

OpenClaw adds Code Mode for tools

This post is a product update about OpenClaw, highlighting a new Code Mode that expands the agent beyond plain shell execution. The core idea is that shell can compose shell commands, but Code Mode can also orchestrate MCP tools, app tools, dynamic tools, structured JSON outputs, loops, filters, maps, and conditionals inside a single runtime cell. The author frames it as an additional composition path rather than a replacement, with an opt-out via `code_mode_only` so the model can still use normal tools directly.

BENCHMARK1h ago

llama.cpp Vulkan stumbles on Arrow Lake

A Reddit user says llama.cpp’s Vulkan backend performs terribly on an Arrow Lake Arc 130T iGPU, with decent prompt processing but sub-4 tok/s generation on Gemma 4 E4B. The thread frames SYCL and other Intel-native backends as the real alternative, not Vulkan.