# Grok 4.5 Launches July 9 as xAI's Opus-Class > Source: [https://icharles.com/articles/grok-4-5-xai-opus-class-launch](https://icharles.com/articles/grok-4-5-xai-opus-class-launch) (canonical) > Author: iCharles News — iCharles, https://icharles.com > Published: 2026-07-08 ## TL;DR xAI's Grok 4.5 went public on July 9, 2026, one day after Elon Musk announced it on X. Built on xAI's new V9 foundation at roughly 1.5 trillion parameters — about 3× the V8-small architecture in earlier Grok 4 models — it is positioned as an "Opus-class model, but faster, more token-efficient and lower cost." Training was supplemented with Cursor coding data. No official pricing had been confirmed at launch. ## What is Grok 4.5? **Grok 4.5** is xAI's newest large language model, announced by Elon Musk on July 8, 2026 and released to the public the following day. It is the first Grok model built on xAI's new V9 foundation, running at roughly 1.5 trillion parameters. That is about three times the scale of the V8-small architecture that powered earlier Grok 4 variants. Musk described it as an "Opus-class model, but faster, more token-efficient and lower cost" — a direct reference to Anthropic's Claude Opus line. According to [xAI's announcement](https://x.ai/news/grok-4-5), internal evaluations claim Grok 4.5 is "comparable to or better than Claude Opus," though no independent public benchmarks existed at launch. --- ## When did Grok 4.5 launch? Musk posted the announcement on X on July 8, 2026. The public rollout followed on July 9. Before that, the model ran in private beta with engineering teams at SpaceX and Tesla. Musk cited "strong positive feedback from customers in our beta test program" as the reason for moving to public release. As [Mashable reported](https://mashable.com/tech/musk-grok-4-5-launch), the stable version of Grok at the time of announcement was Grok 4.3. Grok 4.5 was originally targeted for late May 2026, meaning the public launch arrived roughly a month behind schedule. --- ## What architecture powers Grok 4.5? Grok 4.5 runs on xAI's V9 foundation. Reported figures put it at approximately 1.5 trillion parameters. For comparison, the V8-small architecture behind earlier Grok 4 builds was about one-third that size. xAI had not published a full spec sheet at launch. The context window was not confirmed, though Grok 4.3 supported 1 million tokens and Grok 4.20 pushed to 2 million. Both figures are worth watching once xAI releases official documentation. --- ## Why was Cursor data used in training? The most concrete training detail xAI disclosed is that Grok 4.5's pipeline was supplemented with data from **Cursor**, the AI coding environment. The goal was to sharpen coding and technical performance — the area where Claude Opus has historically been strongest. This matters for developers building [agentic coding workflows](/articles/meta-bans-claude-codex-distillation) or evaluating models for production code tasks. Training on real-world data from a tool developers actively use is different from generic code scraped from the web, according to the sources. --- ## How does Grok 4.5 compare to other models? Here's where Grok 4.5 sits relative to its direct competitors and its own predecessor, based on what the sources confirm: | Model | Positioning | Parameters | Key detail | |---|---|---|---| | Grok 4.5 | Opus-class, faster, cheaper | ~1.5 trillion (V9) | Cursor coding data; no public pricing at launch | | Grok 4.3 | Current public flagship (at launch) | V8-small | $1.25/M input, $2.50/M output tokens | | Claude Opus | xAI's stated benchmark target | Not disclosed | Anthropic's general-purpose frontier model | | Grok 5 | Upcoming flagship, still in training | ~6 trillion (MoE) | Not yet released as of mid-2026 | Grok 4.5 is explicitly an interim step. **Grok 5** — a separate, much larger Mixture-of-Experts model at roughly 6 trillion parameters — was still in training as of the Grok 4.5 launch. --- ## What does Grok 4.5 cost? xAI had not confirmed Grok 4.5 pricing at launch. The reference point is Grok 4.3, which was priced at **$1.25 per million input tokens** and **$2.50 per million output tokens**, with prompt caching around $0.20 per million. Musk's "lower cost" framing suggests Grok 4.5 would land at or below those rates, but that remains unconfirmed. For context, [OpenAI has also been cutting inference costs](/articles/openai-inference-cost-halved-optimization) on key models, making per-token pricing a competitive battleground across the frontier. --- ## Where can you access Grok 4.5? Based on xAI's previous model rollouts, Grok 4.5 is expected to be available through: - **X / Grok app** — for X Premium+ and SuperGrok subscribers - **xAI API** — for developers, priced per token (rates pending official confirmation) - **xAI's developer stack** — including Grok Build, Skills, and Connectors for agentic coding use cases --- ## How does Grok 4.5 fit into xAI's model history? Here's what we know so far about xAI's release cadence leading up to Grok 4.5: - **Grok-1** (Nov 2023) — First beta, X Premium only - **Grok-2** (Aug 2024) — Better reasoning, tool use, vision - **Grok-3** (Feb 2025) — Large jump in reasoning - **Grok-4** (Jul 2025) — Flagship with native tool-calling and real-time search - **Grok 4.1** (Nov 2025) — 65% hallucination reduction - **Grok 4.20** (Feb 2026) — Multi-agent system, 2M context, lower pricing - **Grok 4.3** (Apr 2026) — Faster reasoning, public flagship - **Grok 4.5** (Jul 2026) — V9 foundation, ~1.5T parameters, coding focus The point-releases between major versions are where xAI tightens accuracy, speed, and cost. Grok 4.5 is the largest of those steps yet because it replaces the underlying foundation model rather than tuning the previous one. --- The next confirmed milestone is Grok 5, the ~6-trillion-parameter Mixture-of-Experts model that was still in training as of the Grok 4.5 public launch on July 9, 2026. Official pricing for Grok 4.5 had not been released at that time. --- ## Frequently asked questions **When was Grok 4.5 released to the public?** Elon Musk announced Grok 4.5 on July 8, 2026. The public rollout began the following day, July 9, 2026. Before that, the model ran in a private beta limited to engineering teams at SpaceX and Tesla. The model had originally been targeted for late May 2026, making the public launch about a month behind its original schedule. **What does "Opus-class" mean for Grok 4.5?** Musk used "Opus-class" to position Grok 4.5 as comparable in capability to Anthropic's Claude Opus. xAI's internal evaluations claim it is "comparable to or better than Claude Opus" on quality, while being faster, more token-efficient, and lower cost. No independent public benchmarks had verified that claim at the time of launch. **What is the V9 foundation in Grok 4.5?** The V9 foundation is xAI's new base architecture for Grok 4.5, running at roughly 1.5 trillion parameters. It replaces the V8-small architecture used in earlier Grok 4 models. The V9 foundation is approximately three times the scale of V8-small, which xAI says points to stronger reasoning and broader knowledge capacity. **Is Grok 4.5 the same as Grok 5?** No. Grok 4.5 is an interim upgrade between Grok 4.3 and the upcoming Grok 5. Grok 5 is a separate, much larger model planned at roughly 6 trillion parameters using a Mixture-of-Experts architecture. Grok 5 was still in training as of the Grok 4.5 public launch in July 2026 and had not been released. **How much does Grok 4.5 cost per token?** xAI had not confirmed official pricing for Grok 4.5 at launch. The closest reference is Grok 4.3, priced at $1.25 per million input tokens and $2.50 per million output tokens, with prompt caching at around $0.20 per million. Musk described Grok 4.5 as "lower cost," suggesting pricing at or below Grok 4.3 rates. ## Frequently asked questions **When was Grok 4.5 released to the public?** Elon Musk announced Grok 4.5 on July 8, 2026. The public rollout began the following day, July 9, 2026. Before that, the model ran in a private beta limited to engineering teams at SpaceX and Tesla. The model had originally been targeted for late May 2026, making the public launch about a month behind its original schedule. **What does "Opus-class" mean for Grok 4.5?** Musk used "Opus-class" to position Grok 4.5 as comparable in capability to Anthropic's Claude Opus. xAI's internal evaluations claim it is "comparable to or better than Claude Opus" on quality, while being faster, more token-efficient, and lower cost. No independent public benchmarks had verified that claim at the time of launch. **What is the V9 foundation in Grok 4.5?** The V9 foundation is xAI's new base architecture for Grok 4.5, running at roughly 1.5 trillion parameters. It replaces the V8-small architecture used in earlier Grok 4 models. The V9 foundation is approximately three times the scale of V8-small, which xAI says points to stronger reasoning and broader knowledge capacity. **Is Grok 4.5 the same as Grok 5?** No. Grok 4.5 is an interim upgrade between Grok 4.3 and the upcoming Grok 5. Grok 5 is a separate, much larger model planned at roughly 6 trillion parameters using a Mixture-of-Experts architecture. Grok 5 was still in training as of the Grok 4.5 public launch in July 2026 and had not been released. **How much does Grok 4.5 cost per token?** xAI had not confirmed official pricing for Grok 4.5 at launch. The closest reference is Grok 4.3, priced at $1.25 per million input tokens and $2.50 per million output tokens, with prompt caching at around $0.20 per million. Musk described Grok 4.5 as "lower cost," suggesting pricing at or below Grok 4.3 rates.