Grok Build 0.1

xAI announced on May 29, 2026 that its Grok Build 0.1 coding model (model key grok-build-0.1) is now available via the xAI API in public beta. The model is described as being specifically trained for agentic coding tasks, including web development, debugging, and Model Context Protocol (MCP) support, and as the same mo According to the announcement, grok-build-0.1 is positioned as both a fast coding model and a speedy, economical option for general-purpose agentic and tool-calling workloads beyond coding. The first-party page recommends running it inside agentic harnesses such as Grok Build, Cursor, Hermes Agent, OpenClaw, Kilo Code,

The official xAI news index lists a September 16, 2026 product update titled "Memory in Grok Build," indicating a general-availability memory capability landing inside the Grok Build product line. This is the same feature flagged GA in the Grok Build CLI changelog v1.0.34 on the same date, pointing to a coordinated cro An earlier August 19, 2026 entry, "Grok Build on web and mobile," expanded Grok Build availability to every plan via web and mobile. The page does not explicitly name the grok-build-0.1 model variant in the supplied excerpt, so the entry is treated as family/product-level news directly relevant to the Grok Build experi

The official Grok Build changelog (v1.0.34, September 16, 2026) ships Memory as generally available and fixes markdown heading color theming. It is the most recent and technically detailed first-party update in the batch and directly affects what the Grok Build experience can do, although the page describes CLI/client Earlier entries in v1.0.33–v1.0.31 add structured JSON output from MCP tool results, fix subagent status, MCP permission sheet quirks, and long-session compaction/history retention. The changelog does not explicitly name the grok-build-0.1 model variant, so acceptance is scoped to Grok Build CLI tooling adjacent to the

The supplied xAI release notes describe Grok Build CLI and other Grok family updates, but the excerpt does not explicitly name Grok Build 0.1 or establish a model-side change to that exact variant. Entries such as Grok 4.6, Grok Bot, and Imagine updates are distinct products or models, so their specifications and relea As a result, the excerpt provides no sufficiently specific, timely evidence about the Grok Build 0.1 model itself. It is rejected for exact-version mismatch and for mixing CLI or sibling-model news with the subject model.

The article is about Grok 4.5, a distinct sibling model, and says that Grok 4.5 became the default in Grok Build. Its specifications, pricing, benchmark comparisons, and launch details therefore cannot be transferred to Grok Build 0.1. The candidate is rejected for exact-version mismatch and because it concerns a different model rather than the subject variant.

Grok Build 0.1 was released by xAI on May 29, 2026 in public beta as a coding model with a specific purpose: to function as an autonomous software engineering agent rather than a code-completion helper. The model is the exact same one powering the Grok Build CLI, xAI's agentic command-line tool originally introduced in The underlying weights for Grok Build 0.1 were trained from scratch on a corpus heavily weighted toward real programming content such as pull requests, open-source repositories, debugging sessions, and documentation, meaning the model is purpose-built for code rather than adapted from a general-purpose foundation. Tech

An independent analysis dated June 5, 2026 frames grok-build-0.1 as xAI's bet on parallel breadth in the terminal coding agent arena, noting that xAI released the Grok Build CLI on May 25 in early beta (limited to SuperGrok and X Premium+ subscribers) and opened the underlying grok-build-0.1 model to public beta via th The analysis contrasts grok-build-0.1's eight parallel sub-agents (each in its own worktree) with Claude Code's single deep agent using a 1M-token context window, citing Morph's observation that the trade-off is context — each of the eight agents operates with its own context window rather than sharing a single large o

According to DevOps.com's June 2026 write-up, xAI made Grok Build 0.1, described as xAI's fastest coding model, available via the xAI API in public beta, giving developers API access to the same model that powers the Grok Build CLI without requiring a SuperGrok or X Premium+ subscription. The article frames the API ope The same report details Grok Build 0.1's technical profile: a coding model specifically trained for agentic tasks including web development, debugging, and MCP support; text and image inputs with text output; a 256,000-token context window; support for function calling, structured outputs, and built-in reasoning that i

Independent analysis by Can Demir published May 27, 2026 traces xAI's May 14, 2026 beta launch of Grok Build, the terminal-native coding CLI built on the grok-build-0.1 model and written in Rust. The piece flags that for individual developers Grok Build is gated behind a SuperGrok Heavy subscription at roughly $299/mon The article cross-checks launch coverage against xAI documentation, Anthropic's model docs, and OpenAI's GPT-5.5 announcement, identifying several circulating claims — context window specifics, Plan Mode defaults, Windows support uniqueness, and the model's true identity — that contradict primary sources. It explicitly

Kilo Code's third-party evaluation page for SpaceXAI's Grok Build 0.1 (released May 20, 2026) explicitly names the model and reports a 256,000-token context window with a 230,400-token max output. The page lists the listed input price at $1.00 per 1M tokens and describes the model as SpaceXAI's fast coding model traine On Kilo Bench Terminal Bench 2.0, the model achieves 50.6% completion at $30.70 per attempt, while on the PinchBench OpenClaw suite it averages 88.9% across 7 of 50 official models with a 220m 40s average runtime and $20.581 per benchmark run. Category breakdown shows strong results in Log Analysis (97.0%), CSV Analysi

OpenRouter's model listing for SpaceXAI's grok-build-0.1 (slug x-ai/grok-build-0.1-20260520, released May 20, 2026) describes it as a fast coding model trained specifically for agentic software engineering workflows, supporting text and image inputs with text output and optimized for interactive coding agents, tool use However, OpenRouter is a gateway/router rather than the model creator, and the page also includes hosting-side metrics such as throughput (96 tok/s P50) and latency (0.53s) for the SpaceXAI hosting endpoint. These inference-side performance figures and routing-mode descriptions reflect serving-provider characteristics