xAI

Key Terms

• Grok Build 0.1 - xAI's dedicated coding model, trained for agentic software tasks (web dev, debugging, MCP support). Served at 100+ tokens/second with a 256K context window. Model id grok-build-0.1; aliases grok-code-fast-1, grok-code-fast. Source: xAI – Grok Build 0.1
• Grok Build CLI - xAI's terminal coding agent (command grok), launched May 25 in beta for SuperGrok and X Premium Plus subscribers. Features plan mode, inline diffs, parallel subagents, worktree integrations, MCP server support, and headless mode (-p). Source: xAI – Overview
• SuperGrok - xAI's premium consumer subscription at $30/mo, granting Grok 4, higher rate limits, image/video generation, and Grok Build CLI access. Source: xAI – Pricing
• Cached input - frequently reused prompt tokens stored on xAI's servers at $0.20/MTok, an 84% discount versus standard input ($1.25 on Grok 4.3/4.20). Available on all text models. Source: xAI – Models
• V9 foundation model - the next-generation base model Musk cited as the underpinning of Grok 4.5 ("our 1.5T V9 foundation model"), described as trained from scratch. No technical details (parameter count methodology, training data) have been published. Source: News – Item
• Intelligence Index - Artificial Analysis's composite quality score combining several academic and coding benchmarks, normalized across models. Grok Build 0.1 scored 39.8; the top scored model on the same chart is 59.9. Source: Artificialanalysis – Grok Build 0 1 06 16
• Colossus 1 - xAI's GPU supercomputer, also rented to Anthropic under a compute partnership (announced May 6). Source: xAI – Anthropic Compute Partnership

Latest Changes

Changes since the 2026-05 report.

• Grok 4.5 teased, not shipped: On June 28, Elon Musk posted that Grok 4.5 is "based on our 1.5T V9 foundation model, with Cursor data added in", claiming early evals show "performance close to, perhaps exceeding Opus". As of June 30, Grok 4.5 does not appear in the API pricing table and no release date, price, or context window is published. Source: News – Item
• Acquisition (verified May watch-item, now signed): SpaceX converted its April option into a definitive $60 billion all-stock agreement to acquire Cursor (Anysphere) on June 16, 2026, four days after SpaceX's June 12 IPO. Expected to close in Q3 2026. Source: TechCrunch – Spacex To Acquire Cursor For 60B In Stock Days After Blockbuster Ipo
• Cursor data integration signaled: Musk's Grok 4.5 tease ("with Cursor data added in") is the first public confirmation that Cursor's coding telemetry will feed xAI's model training. Source: News – Item
• Composer 2.5 in Grok Build (reported): A HackerNews submission titled "Composer 2.5 is now available in Grok Build", pointing to xAI – Composer 2 5 , appeared June 2, indicating Cursor's proprietary Composer 2.5 model became usable inside the Grok Build CLI. The x.ai page itself was 403-blocked at fetch time, so the detail is sourced from the submission, not the primary page. Source: News – Item
• API rate limit raised: Grok 4.3 and Grok Build 0.1 now show 37 requests/second (roughly 2,220 RPM) on their docs pages, up from the 1,800 RPM reported in May. Tokens-per-minute stays 10,000,000. Source: xAI – Grok 4.3 , xAI – Grok Build 0.1
• New region added: A third API region, us-west-2, is now listed for Grok 4.3 and Grok Build 0.1, alongside the May regions us-east-1 and eu-west-1. Source: xAI – Grok 4.3
• Independent benchmark published: Artificial Analysis published a "Grok Build 0.1: Intelligence, Performance and Price Analysis" page (linked June 24) scoring the model at Intelligence Index 39.8, output speed 80.27 tokens/sec, cost per intelligence-index task $0.21, and TerminalBench v2.1 at 52.06%. Source: Artificialanalysis – Grok Build 0 1 06 16
• Model lineup unchanged: The live models table still lists grok-4.3, grok-4.20-0309-reasoning, grok-4.20-0309-non-reasoning, grok-4.20-multi-agent-0309, and grok-build-0.1 at identical May prices. No Grok 4.5 row exists. Source: xAI – Models
• Knowledge cutoff unchanged: Grok 3 and Grok 4 family models retain a November 2024 knowledge cutoff; realtime data still requires server-side Web/X Search tools. Source: xAI – Models
• Pricing-page verification blocked: The consumer pricing page (xAI – Pricing) and news index (xAI – News) returned HTTP 403 to both WebFetch and curl, so consumer plan prices are carried forward from the May report and could not be re-verified for June. See Transparency Gaps.

Plans

The consumer and business pricing page (xAI – Pricing) was 403-blocked at fetch time, so the table below is carried forward from the 2026-05 report. Any June plan changes could not be verified. Treat as last-confirmed May state.

Source (last verified May 2026): xAI – Pricing and xAI – Business

Notes on plan limits:

• Concrete usage caps (messages/day, requests/hour, tokens/month) are still not published for any consumer tier; tiers are described only as "limited" and "higher". See Transparency Gaps.
• SuperGrok Heavy pricing remains undisclosed on the pricing page. Community reports of ~$300/mo (with a $99 promo) are still unconfirmed by xAI. Source: Reddit – Grok Build Cli Agentscli And The Cli Coding Tool
• A June HackerNews commenter directly compared the SuperGrok Heavy tier to Claude Code, concluding "You'd be better off with Claude Code 5x Max than the 300 USD/month subscription." Source: News – Item

API Pricing

Source: xAI – Models , xAI – Grok 4.3 , xAI – Grok Build 0.1

Notes:

• Rate-limit and region figures are confirmed live for grok-4.3 and grok-build-0.1. The main models table no longer displays RPM/TPM columns, but the per-model pages show 37 RPS and 10M TPM. The May report listed 1,800 RPM (30 RPS), so this is a raise. Source: xAI – Grok 4.3
• Prompt caching is an 84% discount ($0.20 vs. $1.25 input) across all text models. Source: xAI – Models
• Grok Build 0.1 is the cheapest text model at $1.00/$2.00 per MTok, with a reduced 256K context window. Source: xAI – Grok Build 0.1
• No batch API pricing is listed. No tiered or volume-discount API pricing is documented.

Other API pricing (unchanged from May):

Source: xAI – Models

Model Performance / Benchmarks

xAI still publishes no first-party coding benchmarks (no SWE-Bench Verified, Terminal-Bench, LiveCodeBench, or ARC-AGI submissions). The Grok 4.3 model page repeats the unsupported claim that it is "leading the industry in non-hallucination rate, agentic tool calling, and instruction following capabilities" without citing data. Source: xAI – Grok 4.3

The first independent, third-party evaluation of Grok Build 0.1 appeared in June via Artificial Analysis (page dated to the 0616 checkpoint, surfaced on HackerNews June 24):

For comparison, Grok 4.3 (high reasoning) scored an Intelligence Index of 37.58 on the same evaluator, slightly below Grok Build 0.1's 39.8. Source: Artificialanalysis – Grok Build 0 1 06 16

Community benchmarkers were harsher. One HackerNews commenter wrote that Grok Build 0.1 "looks like a worse Deepseek V4 pro. Costs more, dumber, slower and more verbose than most modern Chinese releases." Source: News – Item

Latest News

Grok 4.5 Teased with Cursor Data (June 28, 2026)

Elon Musk posted that Grok 4.5 is "based on our 1.5T V9 foundation model, with Cursor data added in", and that "early evals show performance close to, perhaps exceeding Opus." He also claimed "completely trained from scratch new models will be released by SpaceX every month this year." No release date, price, context window, or benchmark accompanied the post, and Grok 4.5 is absent from the live API table. Source: News – Item

SpaceX Acquires Cursor (June 16, 2026)

SpaceX signed a definitive $60 billion all-stock agreement to acquire Cursor (Anysphere), four days after SpaceX's June 12 IPO. The deal is expected to close in Q3 2026. SpaceX shares then plunged, wiping out roughly $600 billion in market value. This is the corporate foundation for the Grok 4.5 "Cursor data" promise. Sources: TechCrunch – Spacex To Acquire Cursor For 60B In Stock Days After Blockbuster Ipo , Forbes – Spacex Stock Plunge Wipes Out 600 Billion After Cursor Deal Spooks Investors

Composer 2.5 Available in Grok Build (June 2, 2026, reported)

A HackerNews submission titled "Composer 2.5 is now available in Grok Build", linking to xAI – Composer 2 5 , indicates Cursor's proprietary Composer 2.5 coding model became usable inside xAI's Grok Build CLI. This predates the signed acquisition (June 16) and follows SpaceX's earlier April option to buy Cursor. The x.ai page returned HTTP 403 at fetch time, so the claim rests on the submission and could not be confirmed against the primary post. Source: News – Item

Grok Build 0.1 Independent Analysis (June 24, 2026)

Artificial Analysis published an "Intelligence, Performance and Price Analysis" of Grok Build 0.1 (0616 checkpoint), giving it an Intelligence Index of 39.8, TerminalBench v2.1 of 52.06%, and output speed of 80.27 tokens/sec. The HackerNews thread reached 16 points and 18 comments, with sentiment broadly skeptical of the model's competitiveness. Sources: Artificialanalysis – Grok Build 0 1 06 16 , News – Item

Community Signals

Community discussion around xAI's coding tools grew in June (driven by the Grok 4.5 tease and the Grok Build analysis) but remained largely skeptical, focusing on unverified capability claims, usage limits, and the strategic risk of SpaceX's vertical integration.

Grok 4.5 tease thread (11 points, 6 comments)

• Kelteseth, reacting to the "early evals show performance close to, perhaps exceeding Opus" claim: "Typical Elon marketing speech". News – Item
• r0fl, on why the model is not worth trying regardless of quality: "Grok usage limits are too low for me to care to even try it. I pay for codex, antigravity is free, openrouter lets me use Kiki and glm and deepseek for cheap if I need long tasks. At this point a model would have to be mind blowing for me to put in the effort to change over". News – Item
• Havoc, on the open question for adoption: "Pricing? GLM has shown there is huge demand for near SOTA... but if it isn't priced aggressively then it won't see uptake". News – Item

Grok Build 0.1 analysis thread (16 points, 18 comments)

• blenklo, citing the Artificial Analysis coding metric gap: "The coding index is at 51 while the others are at 75". News – Item
• criley2, on relative quality: "This looks like a worse Deepseek V4 pro. Costs more, dumber, slower and more verbose than most modern Chinese releases. Seeing how those Chinese releases are in various states of open source, I would hope a competitor could at least match them." News – Item
• recsv-heredoc, a detailed product critique after real use, concluding on price: "You'd be better off with Claude Code 5x Max than the 300 USD/month subscription." The commenter also flagged UX friction ("worse than Claude Code"), verbose visible chain-of-thought, and harness crashes. News – Item
• vessenes, noting the evaluator ranked Grok Build below Meta's open offering and predicting an aggressive push once the Cursor integration lands: "I noted with interest they rate grok build slightly below Meta's offering(!). I would anticipate we will see aggressive improvements as the Cursor integration comes together, and a hard push on price and quality over the next few months." News – Item
• grim_io, on xAI's competitive trajectory and talent loss: "Since their very brief rise to the top with grok 3, they have been firing their talent and having their execs leave. The result being that their models are falling behind further every day. Now they are behind Chinese open weight models." News – Item

Overall community sentiment

The dominant June themes are: (1) Grok 4.5 claims are treated as marketing until shown in the API with benchmarks; (2) Grok Build 0.1 is viewed as fast and cheap but below frontier and below several Chinese open-weight models on quality; (3) usage limits and the ~$300/mo SuperGrok Heavy tier are seen as adoption blockers; and (4) SpaceX's ownership of Cursor is recognized as a structural advantage (training data plus the IDE) but one that has not yet translated into a better model.

Enterprise Readiness

The consumer pricing and business pages were 403-blocked at fetch time, so enterprise feature verification carries forward from May where noted.

Transparency Gaps