SpaceXAI Releases Grok 4.5 on 8 July 2026
On 8 July 2026, SpaceXAI released Grok 4.5 — its first model built specifically for coding and agentic work. Available immediately through the SpaceXAI console, Grok Build, and Cursor on all subscription plans, the model closes a pricing gap that has previously separated xAI from the frontier tier: at $2 per million input tokens and $6 per million output tokens, Grok 4.5 costs less than half of Opus 4.8 and GPT-5.5 while scoring within benchmark range of both on agentic software engineering tasks. EU access is expected in mid-July 2026 due to local regulatory requirements.
The V9 Architecture and Cursor Training Data
Grok 4.5 is built on xAI's V9 foundation architecture, estimated at approximately 1.5 trillion parameters — a full generation above the V8-series that powered Grok 4.3. The scale shift triples the parameter count while narrowing the training focus: unlike Grok 4.3's general reasoning orientation, the V9 model was trained on datasets spanning coding, science, engineering, and mathematics. Supplemental supervised fine-tuning and reinforcement learning were then applied on real Cursor developer session data — the actual multi-step debugging, iterative refactoring, and tool-use patterns that Cursor developers executed on production codebases.
This training integration distinguishes Grok 4.5 from other frontier coding models. The model did not simply learn from public code repositories on GitHub: it learned from the lived agentic sessions of professional developers using one of the most widely adopted AI coding environments. Grok 4.5 serves at 80 tokens per second with a 500,000-token context window. The context window is halved from Grok 4.3's one million tokens, but 500,000 tokens is consistent with the depth that most coding agent workflows require in practice.
Benchmark Scores: Where Grok 4.5 Lands
On DeepSWE 1.0 — the leading benchmark for agentic software engineering tasks — Grok 4.5 scores 62.0 per cent. Claude Fable (max) leads at 66.1 per cent, and GPT-5.5 (xhigh) scores 64.31 per cent. On the newer DeepSWE 1.1, Grok 4.5 scores 53 per cent, while Fable (max) reaches 70 per cent and GPT-5.5 (xhigh) scores 67 per cent.
On SWE Marathon — a separate pass@1 resolution benchmark — Grok 4.5 leads with 29 per cent, ahead of Opus 4.8 (max) at 26 per cent and Fable (max) at 24 per cent. Artificial Analysis independently ranks Grok 4.5 fourth among 168 models on its Intelligence Index with a score of 54.
The per-task cost figure is more consequential than headline benchmark rank for most teams. Grok 4.5 resolves tasks using an average of 15,954 output tokens. Opus 4.8 (max) averages 67,020 tokens for comparable tasks — 4.2 times as many. Because per-task cost multiplies output tokens by the per-token price, Grok 4.5's token efficiency compounds the pricing advantage significantly. Published estimates put per-task cost at $2.49 for Grok 4.5 in Grok Build versus $5.07 for GPT-5.5 in Codex and $11.80 for Fable 5 in Claude Code.
Cursor Integration and Availability
Grok 4.5 is available in Cursor on all subscription plans — free, Pro, and Business — making it the first third-party model broadly available across all Cursor tiers. The launch reflects the depth of the Cursor-xAI relationship: Cursor's developer session data contributed to the model's fine-tuning, making Grok 4.5's Cursor integration closer to a co-developed model than a standard API partnership. The model is also available directly through the SpaceXAI console API and Grok Build for teams that want direct access without the Cursor layer.
What Grok 4.5 Means for Indian Engineering Teams
The per-task cost advantage is the most immediately actionable factor for Indian development teams. Agentic coding workflows at production scale — automated code review, test generation, large codebase refactoring — are constrained by total cost per task rather than just per-token pricing. At $2.49 per task versus $5.07 for GPT-5.5 and $11.80 for Fable 5, Grok 4.5 reduces the operational cost of frontier-class agentic coding by a factor of two to five. For Indian engineering teams managing tight budgets across large codebases, this changes the economic case for deploying agentic coding at production scale.
Cursor is the dominant AI-native coding environment at many Indian product companies. Grok 4.5's availability on all Cursor tiers means teams already using Cursor can trial the model without any subscription change. The token efficiency advantage matters for high-volume workflows: resolving tasks in roughly 16,000 output tokens rather than 67,000 reduces both cost and latency for multi-step coding pipelines. Teams that previously evaluated but rejected Fable 5 or GPT-5.5 on cost grounds should reassess with Grok 4.5's per-task pricing in the picture.
The Bottom Line
SpaceXAI released Grok 4.5 on 8 July 2026 — a 1.5-trillion-parameter V9 model fine-tuned on real Cursor developer session data, priced at $2 input and $6 output per million tokens. It scores 62.0 per cent on DeepSWE 1.0 and leads SWE Marathon at 29 per cent pass@1 resolution. Its 4.2x token efficiency advantage over Opus 4.8 puts its per-task coding cost at $2.49 — roughly half GPT-5.5 in Codex and a fifth of Fable 5 in Claude Code. Available in Cursor on all plans and through the SpaceXAI console from launch day. For Indian engineering teams, the practical question is simple: at $2.49 per coding task, the economic case for deploying frontier-class AI agents at production scale changes substantially.
Frequently Asked Questions
What is Grok 4.5 and when did SpaceXAI release it?+
SpaceXAI released Grok 4.5 on 8 July 2026 — its first model built specifically for coding and agentic work. Built on the V9 foundation architecture with approximately 1.5 trillion parameters, it was fine-tuned on real Cursor developer session data including multi-step debugging, iterative refactoring, and tool-use patterns. It is available through the SpaceXAI console, Grok Build, and Cursor on all subscription plans. EU access is expected in mid-July 2026.
How does Grok 4.5 pricing compare to GPT-5.5 and Claude Fable 5?+
Grok 4.5 is priced at $2 per million input tokens and $6 per million output tokens. On a per-task cost basis — which accounts for the number of output tokens each model uses to resolve a coding task — Grok 4.5 costs $2.49 per task versus $5.07 for GPT-5.5 in Codex and $11.80 for Fable 5 in Claude Code. This reflects Grok 4.5's token efficiency: it resolves tasks in an average of 15,954 output tokens, compared to 67,020 for Opus 4.8 (max) — 4.2 times fewer.
What are Grok 4.5's benchmark scores on DeepSWE and SWE Marathon?+
On DeepSWE 1.0, Grok 4.5 scores 62.0 per cent, behind Claude Fable (max) at 66.1 per cent and GPT-5.5 (xhigh) at 64.31 per cent. On DeepSWE 1.1, Grok 4.5 scores 53 per cent versus Fable (max) at 70 per cent and GPT-5.5 at 67 per cent. On SWE Marathon (pass@1 resolution), Grok 4.5 leads at 29 per cent, ahead of Opus 4.8 (max) at 26 per cent and Fable (max) at 24 per cent. Artificial Analysis ranks Grok 4.5 fourth among 168 models on its Intelligence Index with a score of 54.
What context window does Grok 4.5 have and how does it compare to Grok 4.3?+
Grok 4.5 has a 500,000-token context window — half the one-million-token context Grok 4.3 offered. SpaceXAI serves the model at 80 tokens per second. The context window reduction reflects the V9 model's narrowed training focus on coding and agentic tasks rather than general long-context reasoning. For most coding agent workflows, 500,000 tokens is consistent with the context depth required. Teams working on very large monorepos that relied on Grok 4.3's one-million-token window should evaluate whether 500,000 tokens covers their typical session context before upgrading.
Written by
TechPillow Team
Sharing insights on technology, product development, and the Indian tech ecosystem.