Grok 4.5

xAI (branded as SpaceXAI on the launch page) introduced Grok 4.5 on July 16, 2026, describing it as its smartest model built for coding, agentic tasks, and knowledge work, and stating it was trained alongside Cursor. The official launch page reports benchmark scores including DeepSWE 1.0 at 62.0% pass@1, SWE Marathon a The launch page positions Grok 4.5 as focused on real-world engineering excellence, with pricing listed as $2 per million input tokens and $6 per million output tokens, and availability framed through Grok Build as the default coding agent. Competitor benchmark figures are sourced from the respective developers' publis

xAI's official news index documents several Grok 4.5-related entries, including "Grok 4.5 in GitHub Copilot" (Jul 28, 2026), positioning Grok 4.5 as a coding model available inside the GitHub Copilot product. The same index lists earlier and adjacent items such as Grok for Google Workspace and Grok Build updates, indic The index also charts the trajectory beyond 4.5, including the Aug 12, 2026 introduction of Grok 4.6, the Aug 21 Grok 4.6 on Amazon Bedrock listing, the Aug 26 Grok 4.6 on Microsoft Foundry entry, and platform expansions like Google Workspace and Gemini Enterprise Agent Platform. These confirm continued first-party inv

SpaceXAI announced Grok 4.5 on July 8, 2026, positioning it as a broader system for coding, agent workflows, and knowledge-heavy tasks rather than just an X-based chatbot. Elon Musk described the model as "roughly comparable to Opus 4.7, but much faster" and emphasized it was "more token-efficient and lower cost," fram The FullStack retrospective (August 14, 2026) explains that Grok 4.5 runs on a new large-scale foundation model with long-context reasoning, tool use, and stable behavior over real repositories and internal documents, enabling it to read and modify large codebases, coordinate multi-step plans through tools or APIs, and

Bleap's review describes Grok 4.5, released July 8, 2026, as SpaceXAI's flagship mixture-of-experts large language model trained jointly with Cursor for coding, agentic tasks, and knowledge work, with configurable reasoning effort and pricing of $2 per 1M input tokens and $6 per 1M output tokens. It scores 54 on the Ar The article quantifies that advantage by noting a single Intelligence Index task costs about $0.31, roughly 5x cheaper than Claude Sonnet 5 (max), and frames Grok 4.5 as particularly attractive for development, trading, and research tooling where compounded savings matter. The Bleap piece also includes a Grok family hi

SpaceXAI (xAI) released Grok 4.5 on July 8, 2026, positioning it as a frontier model targeted specifically at coding, agentic tasks, and knowledge work. The model was made available immediately through three channels: Grok Build (xAI's coding agent), Cursor, and the xAI developer console via API. The page reports the m Distribution rolled out in stages: EU API users gained access on July 17, 2026, followed by consumer surfaces (Grok.com, X, iOS, Android) on July 22, 2026. The launch narrative emphasized a developer-first strategy, with xAI noting Cursor helped train the model on trillions of tokens from developer and agent interactio

Chatsmith's overview positions Grok 4.5 as xAI's latest flagship Mixture-of-Experts model built for engineering, coding, and complex reasoning, with training emphasizing software engineering, mathematics, physics, science, technical documentation, and real-world coding repositories. The review frames Grok 4.5 as a pivo The piece compares Grok 4.5 against GPT-5.6, Claude Sonnet 5, and Gemini for workflow fit, claiming stronger software engineering capabilities, more reliable reasoning, and faster responses than prior Grok versions while keeping operating costs lower than many competing flagships. It is positioned as a candidate for te

LLM Stats tracks Grok 4.5 with a composite LLM Stats Score placing it at rank 30 on their leaderboard, with a blended price of $2.19 per million tokens. The page records specific benchmark results sourced from official scorecards, including an AA-Omniscience Index score of 26 (normalized to 126.00/200, rank 1) and a GP Performance-by-category breakdowns show Grok 4.5 performing strongest in Chat (+1.25σ across 69 evaluations) and Websites (+1.39σ across 16), with moderate strength in SVG, 3D, and Animations, and a slight underperformance in Games (−0.38σ). The page serves as a composite benchmark aggregator for Grok 4.5 drawn from th

DataCamp provides a detailed technical review of Grok 4.5, confirming key specifications: 500,000-token context window, knowledge cutoff of February 1, 2026, multimodal input (text and images) with text-only output, and adjustable reasoning effort (low, medium, or high with high as default). The model is SpaceXAI's pro The review documents Cursor's claim of jointly training Grok 4.5 on trillions of tokens from developer and agent interactions, and examines xAI's benchmark charts showing mixed results. DataCamp independently tested the model through the API and Cursor, analyzing pricing, access routes, and cost-per-completed-task metr

A third-party breakdown of pre-launch signals for Grok 4.5 documents a canary UI trace reading "Unlock the full power of Chat with Grok 4.5" that surfaced in the Grok web UI on July 6, 2026, alongside reports of a 1.5T-parameter V9 foundation model with Cursor SFT/RL data added during supplemental training. The article For developers interested in the provenance and training lineage of Grok 4.5, the article compiles the concrete data points that circulated before launch, including parameter count estimates (1.5T V9, with a separate reported 2T V-series run expected to finish in late July) and the Cursor training-data mix. All claims

Released July 8, 2026, Grok 4.5 is described by Value Add as scoring 54 on the Artificial Analysis Intelligence Index (behind Claude Fable 5 at 60, Opus 4.8 at 56, and GPT-5.5 at 55) while leading on real professional work tasks with a 29% pass rate versus 22% for GPT-5.5 and 21% for Opus 4.8. The model is reported to The Value Add analysis emphasizes cost efficiency, citing $0.077 per benchmark point, which it frames as 2.4x cheaper than Opus 4.6 on that basis, and reports Grok monthly active users at 117M as of March 2026 (up from 35M in December 2025). It positions Grok 4.5 as a practical fit for paying customers prioritizing pro

According to LLM Rumors' July 10, 2026 analysis, xAI's Grok 4.5 climbed 16 points on the Artificial Analysis Intelligence Index, moving from 38 (Grok 4.3) to 54 in a single release, putting it within six points of the displayed leader. The piece also cites Artificial Analysis measurements of roughly 89.5–90 output toke The same report positions Grok 4.5 for Cursor-trained agent behavior, live search, and competitive cost rather than a head-to-head two-company contest, arguing the frontier field is now too broad. It describes xAI as having "moved from the lower edge of the displayed frontier field to within six points of its leader" w

A July 9, 2026 Medium recap by Mayur Jain (MLWorks) confirms Grok 4.5 launched on July 8, 2026 as SpaceXAI's smartest model, built to excel at coding, agentic tasks, and knowledge work. It was trained alongside Cursor and sits on the V9 foundation architecture — described as a ground-up redesign of xAI's model foundati On the Artificial Analysis Intelligence Index, Grok 4.5 scores 54, placing fourth behind Fable 5, GPT-5.5, and Opus 4.8. On the Artificial Analysis Coding Agent Index, Grok 4.5 scores on par with GPT-5.5 in Codex while running in the Grok Build harness, and does so at much lower cost. The piece is a secondary synthesis