Grok 4.5 in Microsoft Office: the Vendor Map Shifts

On July 8, 2026, xAI launched Grok 4.5, a mixture-of-experts model trained on real coding data from Cursor across tens of thousands of NVIDIA GB300 GPUs, priced at $2 per million input tokens and $6 per million output tokenswith a 500K-token context window. Its immediate deployment as the default model in Microsoft Word, Excel, and PowerPoint makes it the first AI model from outside the Microsoft portfolio to ship as the default in the Office suite.

Grok 4.5 enters a market where enterprise AI procurement decisions had been consolidating around a small set of major providers. The model's position as default in Office add-ins, reaching hundreds of millions of enterprise users, combined with integrations across Databricks Mosaic, SnowflakeOpenRouter, Vercel, and Cloudflare, creates a distribution footprint that bypasses the typical enterprise sales cycle entirely.

Performance Profile: Engineering, Legal, and Knowledge Work

xAI reports Grok 4.5 achieves a 64.7% resolve rate on SWE-Bench Proresolving tasks with an average of 15,954 output tokens, approximately 4.2× fewer output tokens than Claude Opus 4.8 (max) on the same benchmark. The model ranks first on Harvey's Legal Agent Benchmark, with documented capabilities across multi-sheet Excel modeling with live web research, structured document writing in Word, and diagram creation in PowerPoint. Terminal Bench 2.1 shows an 83.3% score; DeepSWE 1.0 reaches 62.0%. The model operates at 80 tokens per second serving speed.

Training ran across tens of thousands of NVIDIA GB300 GPUs on datasets spanning coding, science, engineering, and mathematics, with reinforcement learning covering hundreds of thousands of multi-step software engineering tasks.

Distribution Beyond the Sales Cycle

The strategic significance of the Office integration is distribution velocity. Grok 4.5 reaches enterprise employees as a pre-installed default in tools already licensed and deployed, bypassing the approval gates that govern traditional AI procurement. For AI vendors competing for enterprise mindshare, a model already in the productivity suite has a deployment advantage that dedicated sales processes cannot replicate.

Free usage for a limited period in Grok Build and Cursor lowers the adoption barrier further. EU availability is expected in mid-July 2026, a timeline CFOs and CDOs at European organizations should incorporate into Q3 vendor evaluation planning.

What the Procurement Decision Looks Like Now

For Chief Digital Officers and CFOs reviewing AI vendor portfolios, Grok 4.5's entry introduces a pricing and distribution pressure that merits reassessment of existing commitments. At $2/M input tokens with a 500K context window, it creates a cost-per-capability benchmark that other providers will respond to. Its presence in Office means many organizations are already running it, ahead of formal procurement approval.

The actionable board decision: establish a monitoring posture on Grok 4.5 EU availability (expected mid-July), assess whether Office add-in usage constitutes a shadow AI procurement event requiring governance review, and evaluate the Harvey Legal benchmark results against current legal tech AI spend. The vendor map for enterprise AI shifted on July 8, the question is whether your organization's AI governance framework reflects the new configuration.