SpaceXAI ships Grok 4.5, the first joint model with Cursor, at half the price of Opus 4.7
Thursday's release is the first frontier model since SpaceXAI closed the $60 billion Cursor acquisition on June 16. Grok 4.5 runs on a 1.5-trillion-parameter MoE trained alongside the Cursor engineering harness, prices at $2 input and $6 output per million tokens, and lands in Grok Build, the SpaceXAI console and every Cursor plan on day one. Elon Musk called it an "Opus-class model, but faster, more token-efficient and lower cost" on a live stream at 09:00 PT. Public availability follows on July 9. The European Union is excluded from the launch pending Digital Services Act review.
Bloomberg · TechCrunch · Axios · Forbes · 2026-07-08 · Source ↗ ·
SpaceXAI released Grok 4.5 on Thursday morning at a live-streamed event from the Palo Alto Cursor office. Bloomberg's Rachel Metz broke the pricing terms during the stream, and TechCrunch's Kyle Wiggers and Axios's Ina Fried both matched with independent confirmations within thirty minutes. Grok 4.5 is the first frontier model shipped by the combined company since SpaceX closed its $60 billion all-stock acquisition of Cursor on June 16, and the model is trained jointly with the Cursor coding harness for tight in-editor integration.
The model runs on SpaceXAI's V9 base architecture at 1.5 trillion parameters in a mixture-of-experts configuration. Pricing lands at $2 per million input tokens and $6 per million output tokens. That undercuts Anthropic's Claude Opus 4.7 by 60 percent on input and 76 percent on output, and undercuts OpenAI's GPT-5.6 Sol by 60 percent on input and 80 percent on output. Musk framed the release as "an Opus-class model, but faster, more token-efficient and lower cost" and said internal benchmarks put Grok 4.5 "roughly comparable to Opus 4.7, but much faster."
Grok 4.5 was trained with Cursor from the first checkpoint. It is the first model in the world that was designed as an IDE tenant, not a chat tenant.— Elon Musk, SpaceXAI CEO, on the launch live stream
Distribution and the Cursor merge
Day-one availability covers three surfaces: the Grok Build API console, the SpaceXAI web console, and the Cursor editor across every subscription tier including the free plan. Cursor Pro and Cursor Business customers get an eight-month usage allowance at pre-acquisition Cursor pricing while the combined billing stack migrates onto SpaceXAI. The European Union is excluded from launch pending Digital Services Act review; SpaceXAI told The Next Web the model will "arrive when European regulators complete their capability assessment." Public consumer availability at grok.com follows on July 9.
The Cursor acquisition at $60 billion was announced on June 16 and closed inside 22 days, unusually fast for a deal of that size. Cursor co-founder and CEO Michael Truell moves into a President of AI Product role reporting directly to Musk. The acquisition was priced roughly 3x Cursor's last venture mark and about 45x annualised revenue, which puts the transaction inside the same private-market range as Meta's Scale AI purchase in June 2025 and Google's Anthropic Series H participation last November.
Benchmarks and positioning
SpaceXAI published benchmark scores framing Grok 4.5 as "just short of best-in-class" against the current frontier. The company shared internal numbers for coding, terminal use and multi-turn agent work, though independent verification is still pending. Forbes noted Musk's line that "the token efficiency is twice greater than the nearest competing model," which SpaceXAI attributes to a new routing scheme in the MoE that assigns short-context tokens to smaller expert clusters. Cybernews framed the release as the first head-to-head competitor to Claude at the specific price band Anthropic dominates for agentic coding.
The token efficiency is twice greater than the nearest competing model.— Elon Musk, cited by Forbes
Market context
The launch runs into a busy Thursday. OpenAI confirmed the same morning that GPT-5.6 Sol, Terra and Luna would exit the U.S. government's restricted-access window and go public on Thursday afternoon Pacific time. OpenAI also rolled out GPT-Live-1 and GPT-Live-1 mini voice models globally. Google's Gemini 3.5 Pro release slipped to July 17 after a full architectural rebuild off the 2.5 Pro base. Anthropic pushed Claude Cowork to web and mobile with new Microsoft 365 write tools. The four-frontier release day is the tightest overlap since November 2024.
The pricing move is the sharpest part of the story. Grok 4.5 sits at the middle tier of the current frontier stack in raw capability, but at pricing that Anthropic has historically reserved for one tier down. Anthropic's Sonnet 5, launched on June 30, prices at $2 input and $10 output through August 31 on a promotional rate. Grok 4.5 undercuts even that promotional price on output tokens. For agentic coding workloads where output volume dominates the bill, the shift is material.
SpaceXAI made a straightforward move today: it took the Anthropic price band and undercut it while claiming Anthropic-tier capability.— Kyle Wiggers, TechCrunch
What the release means for Cursor
Cursor's product pitch since 2023 has been that the model matters less than the harness. The Grok 4.5 launch is the first material test of that thesis under joint ownership. Cursor now ships a model that other harnesses cannot match, since the model is trained on IDE session data no third party has. Windsurf, Aider and Zed all have Cursor-compatible product surfaces but cannot access the joint training data. Cognition, which competes with Devin through its own harness plus Windsurf acquisition, told Axios that "the harness gap is exactly what we have been building around."
The next test is whether the EU delay narrows. If the Digital Services Act review returns a positive assessment inside eight weeks, Grok 4.5 sits inside every major legal market before Google DeepMind ships Gemini 3.5 Pro on July 17. If the review stretches to Q4, SpaceXAI cedes the entire European enterprise coding market to Anthropic and OpenAI through year-end.
SpaceXAI ships Grok 4.5, the first joint model with Cursor, at half the price of Opus 4.7
Thursday's release is the first frontier model since SpaceXAI closed the $60 billion Cursor acquisition on June 16. Grok 4.5 runs on a 1.5-trillion-parameter MoE trained alongside the Cursor engineering harness, prices at $2 input and $6 output per million tokens, and lands in Grok Build, the SpaceXAI console and every Cursor plan on day one. Elon Musk called it an "Opus-class model, but faster, more token-efficient and lower cost" on a live stream at 09:00 PT. Public availability follows on July 9. The European Union is excluded from the launch pending Digital Services Act review.
Bloomberg · TechCrunch · Axios · Forbes · 2026-07-08 · Source ↗ ·
SpaceXAI released Grok 4.5 on Thursday morning at a live-streamed event from the Palo Alto Cursor office. Bloomberg's Rachel Metz broke the pricing terms during the stream, and TechCrunch's Kyle Wiggers and Axios's Ina Fried both matched with independent confirmations within thirty minutes. Grok 4.5 is the first frontier model shipped by the combined company since SpaceX closed its $60 billion all-stock acquisition of Cursor on June 16, and the model is trained jointly with the Cursor coding harness for tight in-editor integration.
The model runs on SpaceXAI's V9 base architecture at 1.5 trillion parameters in a mixture-of-experts configuration. Pricing lands at $2 per million input tokens and $6 per million output tokens. That undercuts Anthropic's Claude Opus 4.7 by 60 percent on input and 76 percent on output, and undercuts OpenAI's GPT-5.6 Sol by 60 percent on input and 80 percent on output. Musk framed the release as "an Opus-class model, but faster, more token-efficient and lower cost" and said internal benchmarks put Grok 4.5 "roughly comparable to Opus 4.7, but much faster."
Grok 4.5 was trained with Cursor from the first checkpoint. It is the first model in the world that was designed as an IDE tenant, not a chat tenant.— Elon Musk, SpaceXAI CEO, on the launch live stream
Distribution and the Cursor merge
Day-one availability covers three surfaces: the Grok Build API console, the SpaceXAI web console, and the Cursor editor across every subscription tier including the free plan. Cursor Pro and Cursor Business customers get an eight-month usage allowance at pre-acquisition Cursor pricing while the combined billing stack migrates onto SpaceXAI. The European Union is excluded from launch pending Digital Services Act review; SpaceXAI told The Next Web the model will "arrive when European regulators complete their capability assessment." Public consumer availability at grok.com follows on July 9.
The Cursor acquisition at $60 billion was announced on June 16 and closed inside 22 days, unusually fast for a deal of that size. Cursor co-founder and CEO Michael Truell moves into a President of AI Product role reporting directly to Musk. The acquisition was priced roughly 3x Cursor's last venture mark and about 45x annualised revenue, which puts the transaction inside the same private-market range as Meta's Scale AI purchase in June 2025 and Google's Anthropic Series H participation last November.
Benchmarks and positioning
SpaceXAI published benchmark scores framing Grok 4.5 as "just short of best-in-class" against the current frontier. The company shared internal numbers for coding, terminal use and multi-turn agent work, though independent verification is still pending. Forbes noted Musk's line that "the token efficiency is twice greater than the nearest competing model," which SpaceXAI attributes to a new routing scheme in the MoE that assigns short-context tokens to smaller expert clusters. Cybernews framed the release as the first head-to-head competitor to Claude at the specific price band Anthropic dominates for agentic coding.
The token efficiency is twice greater than the nearest competing model.— Elon Musk, cited by Forbes
Market context
The launch runs into a busy Thursday. OpenAI confirmed the same morning that GPT-5.6 Sol, Terra and Luna would exit the U.S. government's restricted-access window and go public on Thursday afternoon Pacific time. OpenAI also rolled out GPT-Live-1 and GPT-Live-1 mini voice models globally. Google's Gemini 3.5 Pro release slipped to July 17 after a full architectural rebuild off the 2.5 Pro base. Anthropic pushed Claude Cowork to web and mobile with new Microsoft 365 write tools. The four-frontier release day is the tightest overlap since November 2024.
The pricing move is the sharpest part of the story. Grok 4.5 sits at the middle tier of the current frontier stack in raw capability, but at pricing that Anthropic has historically reserved for one tier down. Anthropic's Sonnet 5, launched on June 30, prices at $2 input and $10 output through August 31 on a promotional rate. Grok 4.5 undercuts even that promotional price on output tokens. For agentic coding workloads where output volume dominates the bill, the shift is material.
SpaceXAI made a straightforward move today: it took the Anthropic price band and undercut it while claiming Anthropic-tier capability.— Kyle Wiggers, TechCrunch
What the release means for Cursor
Cursor's product pitch since 2023 has been that the model matters less than the harness. The Grok 4.5 launch is the first material test of that thesis under joint ownership. Cursor now ships a model that other harnesses cannot match, since the model is trained on IDE session data no third party has. Windsurf, Aider and Zed all have Cursor-compatible product surfaces but cannot access the joint training data. Cognition, which competes with Devin through its own harness plus Windsurf acquisition, told Axios that "the harness gap is exactly what we have been building around."
The next test is whether the EU delay narrows. If the Digital Services Act review returns a positive assessment inside eight weeks, Grok 4.5 sits inside every major legal market before Google DeepMind ships Gemini 3.5 Pro on July 17. If the review stretches to Q4, SpaceXAI cedes the entire European enterprise coding market to Anthropic and OpenAI through year-end.