xAI launches Grok Build 0.1 API with $1/$2 pricing and 256K context

• WesRoth's launch post says xAI put Grok Build 0.1 into the API at $1 per million input tokens and $2 per million output tokens, with a 256K context window.
• Day-one distribution moved fast: OpenCode's availability post, Vercel's AI Gateway post, AskVenice's Venice launch thread, and NousResearch's Hermes Agent post all surfaced the model within about a day.
• Subscription access, not just API access, was part of the rollout, according to OpenCode's Grok and X Premium announcement and thdxr's repost of xAI's note about subscription use in OpenCode.
• In kilocode's test thread, Kilo said five small web builds landed between $0.07 and $0.14 per task, including a landing page, dashboard, and browser snake game.

You can jump from Vercel's changelog to OpenCode's v1.15.7 release notes, and Venice even exposed a direct chat entrypoint. The fast part of this launch was not the benchmark discourse, it was how many agent surfaces wired the model in immediately.

The public spec that showed up first was simple: coding model, 256K context, low API pricing.

Vercel's day-one listing also pinned the model slug as xai/grok-build-0.1 in AI Gateway via vercel_dev's post. Venice added one extra number that did not appear in the initial pricing tweet, with AskVenice's launch thread describing the model as 256K context with 64K output, plus function calling, structured outputs, vision input, and reasoning.

OpenCode turned the rollout into an actual workflow story instead of a raw API drop.

According to OpenCode's post, Grok and X Premium subscribers can use that subscription directly inside OpenCode. Then LLMpsycho's release-note summary pointed to OpenCode v1.15.7 with Grok OAuth and device login, alongside error-handling fixes and restored OpenAI OAuth in the same update.

That matters for one specific reason: access got bundled into existing coding clients immediately, instead of forcing every early user through a fresh xAI-only toolchain.

Kilo's thread is the most concrete early usage evidence in the pool because it attached dollar figures to finished UI outputs.

Across the five examples in the main Kilo thread, the posted costs were:

• Landing page: $0.08, per kilocode's landing page run
• E-commerce page: $0.07, per kilocode's e-commerce run
• GTA VI-style settings screen: $0.09, per kilocode's settings screen run
• Admin dashboard: $0.12, per kilocode's dashboard run
• Browser snake game: $0.14, per kilocode's snake game run

Those are toy-ish tasks, but they are useful toy-ish tasks. The thread gives a rough first-pass price band for short agentic front-end jobs without inventing a benchmark harness.

The ecosystem story is broader than one client.

Within the evidence window, Grok Build 0.1 appeared in:

• OpenCode, per OpenCode's launch post
• Vercel AI Gateway, per vercel_dev's changelog post
• Venice, which framed it as a private deployment with tool use and vision, per AskVenice's launch thread
• Hermes Agent early access, per NousResearch's post
• Kilo, which layered it into its coding workflow and marketing, per kilocode's Grok x Kilo post

Kilo also said its CLI is built on the MIT-licensed OpenCode project and sends improvements upstream according to kilocode, which helps explain why Grok support showed up across adjacent agent tooling so quickly.

The API launch also leaked a slightly stranger product shape around it.

In superdoteng's demo, Grok Build appears inside a Rust chat UI with collapsible tool calls, live skills, inline diffs, streaming, and a terminal toggle. Separately, WesRoth's reposted screenshot shows native /imagine and /imagine-video commands in the CLI, with the capture displaying an always-approve mode and free-credit meter.

So the first public picture of Grok Build 0.1 is not just a cheap coding endpoint. It is an API model that landed at the same time as a growing cluster of agent clients, OAuth flows, and even media-generation commands around the CLI.