{"slug": "spacexai-releases-grok-4-5-with-cursor-trained-coding-push", "title": "SpaceXAI releases Grok 4.5 with Cursor-trained coding push", "summary": "SpaceXAI released Grok 4.5, a new AI model trained alongside coding tool Cursor, aimed at coding and technical tasks. Cursor co-founder Michael Truell called it an \"Opus-class model\" that is fast and low cost, marking a deeper integration between the two companies. The model is now available in Cursor and SpaceXAI's console, with pricing at $2 per million input tokens and $6 per million output tokens.", "body_md": "[Michael Truell (@mntruell)](https://x.com/mntruell) said Wednesday that [SpaceXAI](https://x.ai/news/grok-4-5) released Grok 4.5, a new model trained alongside Cursor and aimed at coding, agentic work and technical knowledge tasks.\n\nTruell, one of [Cursor's four founders](https://www.accel.com/companies/cursor), framed the release as a step beyond Cursor's own model work. In a [post on X](https://x.com/mntruell/status/2074916251743457787) on July 8th, he called Grok 4.5 an \"Opus-class model\" that is \"fast and low cost\" and said it is a \"significant step up\" over every model Cursor has developed, including Composer 2.5. He added that Grok 4.5 has become the daily driver for many people on Cursor's team.\n\nThat matters because Truell's product has spent the last year proving that distribution in the editor can be as valuable as the model itself. Cursor became a default interface for developers using models from OpenAI, Anthropic and others. With Grok 4.5, Cursor is putting more of the model layer under its own orbit with SpaceXAI rather than relying only on outside frontier labs.\n\nSpaceXAI's [launch page](https://x.ai/news/grok-4-5) says Grok 4.5 is its strongest model to date and was trained alongside [Cursor](https://cursor.com/changelog/composer-2-5). SpaceXAI says the model was trained across tens of thousands of Nvidia GB300 GPUs and is built for coding, science, engineering and math. SpaceXAI also says Grok 4.5 is now available in Grok Build, in Cursor on all plans and through the SpaceXAI console, though SpaceXAI says it is not yet available in the EU and expects EU availability in mid-July.\n\nThe launch is the first clear product result of a deeper Cursor-SpaceXAI training relationship. Cursor's [May 18th Composer 2.5 post](https://cursor.com/blog/composer-2-5) said Cursor and SpaceXAI were training a significantly larger model from scratch using 10 times more total compute, with Colossus 2's million H100-equivalents and Cursor's data and training techniques. Cursor did not call that future model Grok 4.5 at the time. Wednesday's release now gives the partnership a named model, a price sheet and a distribution path through both Cursor and SpaceXAI.\n\nGrok 4.5 lands less than two months after Cursor shipped Composer 2.5. Cursor said on May 18th that Composer 2.5 improved on Composer 2 by scaling training, generating more complex reinforcement-learning environments and using targeted textual feedback. Cursor also said Composer 2.5 was trained with 25 times more synthetic tasks than Composer 2 and was built on the same open-source checkpoint as Moonshot's Kimi K2.5.\n\nThe pricing shows why SpaceXAI and Cursor are pushing Grok 4.5 as a developer model rather than a prestige benchmark release. SpaceXAI's [developer docs](https://docs.x.ai/developers/models/grok-4.5) list Grok 4.5 at $2 per million input tokens and $6 per million output tokens, with a 500,000-token context window. Cursor's Composer 2.5 standard tier is cheaper at $0.50 per million input tokens and $2.50 per million output tokens, while Composer 2.5 Fast is priced at $3 per million input tokens and $15 per million output tokens. Grok 4.5 sits between those two Cursor tiers on posted price, while Truell is claiming it is a major capability upgrade.\n\nSpaceXAI is also making aggressive performance claims. Its launch page says Grok 4.5 scored 62.0% on DeepSWE 1.0, 53% on DeepSWE 1.1, 83.3% on Terminal Bench 2.1 and 64.7% on SWE Bench Pro. Those figures are useful directional signals, but SpaceXAI's own page notes that some comparisons run within each model provider's harness. For buyers, the more testable claim may be cost and latency: SpaceXAI says Grok 4.5 runs at 80 tokens per second and uses fewer output tokens than Opus 4.8 on SWE Bench Pro tasks.\n\nThe founder angle is sharper than the benchmark chart. Truell and his co-founders built Cursor around the premise that the coding interface would become the place where model quality is judged in production. Accel lists Cursor's founders as Truell, Aman Sanger, Arvid Lunnemark and Sualeh Asif, and Axios [reported Wednesday](https://www.axios.com/2026/07/08/spacexai-grok-new-model) that Grok 4.5 is SpaceXAI's first model release since going public and acquiring Cursor. Cursor gives SpaceXAI a workflow where developers can test a model on real codebases all day. SpaceXAI gives Cursor access to training scale that few application companies can finance alone.\n\nThat trade is now visible in product form. Grok 4.5 gives Cursor a stronger answer to developers comparing it with Claude Code, OpenAI's coding tools and GitHub Copilot. It also gives SpaceXAI a way to sell model access through the daily habits of programmers rather than through chatbot subscriptions alone.\n\nThe unresolved part is how much of Grok 4.5's edge comes from SpaceXAI's base model work versus Cursor's developer data, evaluation loops and editor distribution. Neither Truell's post nor SpaceXAI's launch page separates those inputs. The market will test the result in a place benchmark tables cannot control: existing repositories, messy tickets, long-running agent sessions and bills that engineering managers can read line by line.", "url": "https://wpnews.pro/news/spacexai-releases-grok-4-5-with-cursor-trained-coding-push", "canonical_source": "https://runtimewire.com/article/spacexai-grok-45-cursor-michael-truell", "published_at": "2026-07-08 18:46:18+00:00", "updated_at": "2026-07-08 18:57:24.494290+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-products", "ai-tools", "developer-tools"], "entities": ["SpaceXAI", "Cursor", "Grok 4.5", "Michael Truell", "Nvidia", "GB300", "Colossus 2", "Moonshot"], "alternates": {"html": "https://wpnews.pro/news/spacexai-releases-grok-4-5-with-cursor-trained-coding-push", "markdown": "https://wpnews.pro/news/spacexai-releases-grok-4-5-with-cursor-trained-coding-push.md", "text": "https://wpnews.pro/news/spacexai-releases-grok-4-5-with-cursor-trained-coding-push.txt", "jsonld": "https://wpnews.pro/news/spacexai-releases-grok-4-5-with-cursor-trained-coding-push.jsonld"}}