Grok 4.5 Aims Squarely at Opus With Cheaper, Code-First AI

Grok 4.5: An Opus-Class Challenger Defined by Price and Workflow

Grok 4.5 release refers to xAI’s new flagship AI model, built on a V9 foundation architecture to target frontier-level reasoning while undercutting rival pricing, and designed to sit inside real developer and enterprise workflows rather than exist only as a generic chatbot upgrade. Elon Musk framed Grok 4.5 as “an Opus-class model, but faster, more token-efficient and lower cost,” a direct shot at Anthropic’s Claude Opus 4.8. That messaging is backed by numbers: Grok 4.5 is priced at USD 2 (approx. RM9.2) per million input tokens and USD 6 (approx. RM27.6) per million output tokens, versus USD 5 (approx. RM23) and USD 25 (approx. RM115) for Opus 4.8. The model is already live in Grok Build, across every Cursor plan, and in the xAI API console, immediately plugging into coding agents and knowledge work stacks. In short, Grok 4.5 is less about novelty and more about a blunt proposition: frontier-class intelligence at discount-lab rates.

V9 Architecture and Cursor Data: Built to Be a Coding Workhorse

Under the hood, Grok 4.5 runs on a 1.5 trillion-parameter mixture-of-experts foundation model xAI calls V9, and it is the first Grok trained with data from Cursor’s coding assistant. That matters more than marketing slogans: V9 is tuned for multi-step software engineering and technical work, with reinforcement learning that targets long agentic rollouts rather than single-shot question answering. Grok 4.5 supports text and image input, text output, a 500,000-token context window, function calling, structured outputs, web search, X search, code execution, and configurable reasoning effort that can be set to low, medium, or high, though reasoning itself cannot be disabled. SpaceXAI is pitching the model around concrete engineering tasks, and its published benchmarks show 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. Performance claims against Opus and GPT-5.x are still contested, but the intent is clear: make Grok 4.5 the coding engine inside terminals, CLIs, and IDEs rather than a chat window you alt‑tab into.

Pricing War: Grok 4.5 Versus Opus and GPT-5.6

The most disruptive thing about Grok 4.5 is not its architecture; it is its AI model pricing. Base API pricing is USD 2 (approx. RM9.2) per million input tokens, USD 0.50 (approx. RM2.3) per million cached input tokens, and USD 6 (approx. RM27.6) per million output tokens, with higher-context pricing above 200,000 tokens. Musk’s comparison to Opus is deliberate: Claude Opus 4.8 sits at USD 5 (approx. RM23) per million input and USD 25 (approx. RM115) per million output, while a faster Grok 4.5 variant at USD 4 (approx. RM18.4) and USD 18 (approx. RM82.8) still stays below Anthropic’s numbers. GPT-5.6 complicates the landscape, with Sol priced at USD 5 (approx. RM23) and USD 30 (approx. RM138), Terra at USD 2.50 (approx. RM11.5) and USD 15 (approx. RM69), and Luna at USD 1 (approx. RM4.6) and USD 6 (approx. RM27.6). On honest comparisons against frontier tiers like Sol and Opus 4.8, Grok 4.5 is currently the cheapest lab claiming top-tier reasoning. For enterprises burning through millions of tokens a day in coding or document workflows, that delta is not cosmetic—it rewrites the budget envelope for agentic systems.

Voice Agent Builder and New Voices: Turning Grok Into a Phone Agent Platform

If Grok 4.5 is the brain, xAI’s expanded voice stack is the mouth and ears. The company is rolling out 21 new flagship voices for Grok Voice, expanding its built-in roster from five to 26 and targeting multilingual voice agents, text-to-speech, and no-code agent creation. Each of the 25+ flagship voices natively supports 25+ languages, with named options like Carina, Orion, Luna, Iris, Altair, Zenith, and more mapped to use cases such as support, advertising, narration, sales, and audiobooks. The original five voices have been retrained for more natural pacing, phrasing, and emphasis, and the system runs across Grok’s real-time and TTS stack with expressive tags for pauses, laughter, breathing, whispering, pitch, speed, singing, and emphasis. This expansion is tightly coupled to a new Voice Agent Builder, a beta no-code tool that lets users create phone-ready voice agents in about two minutes, with telephony, knowledge retrieval, tools, guardrails, MCP, observability, SIP numbers, call recordings, transcripts, and tool-use logs. xAI states that agents are billed at the API rate of USD 0.05 (approx. RM0.23) per minute of audio, with voice included and telephony on a free-provisioned number at USD 0.01 (approx. RM0.046) per minute.

Enterprise AI Tools: From IDEs and Office to Legal, Finance, and What’s Next

Where Grok 4.5 becomes interesting is in how aggressively xAI is threading it into enterprise AI tools. The model is live in Grok Build, Cursor across all plans, and the API console, with EU access promised for mid-July. It plugs into Office add-ins for Word, PowerPoint, and Excel, where xAI says it can create spreadsheet models, slide diagrams, and document drafts. Beyond coding, the company is pitching Grok 4.5 into legal and financial work—the kind of document-heavy, rule-bound tasks that eat billable hours and lend themselves to careful reasoning. That overlap with Anthropic’s existing enterprise relationships is intentional competition, not coincidence. At the same time, independent testing notes Grok Build as fast but “not beating GPT-5.5 or Opus 4.7 on quality,” positioning Grok 4.5 as a high-speed second opinion instead of an undisputed best-in-class. Whether it can close that qualitative gap will be clear soon enough, since Cursor’s user base will test it head-to-head as Thursday’s wider release lands. The bet is simple: if you give enterprises a coding-focused, voice-enabled, Opus-class model at Luna-ish prices, they will tolerate slightly lower peak quality in exchange for scale.