Grok 4.5 Release: Price, Features and Availability in 2026

Grok 4.5 reached the SpaceXAI API and developer tools in July 2026 before rolling out across grok.com, X, iOS, and Android. It costs $2 per million input tokens and $6 per million output tokens. Here is the verified release timeline, what the model can do, where it is available, and what teams should test before switching.

Quick take: Grok 4.5 is unusually cheap for a frontier model, but the useful story is the rollout. SpaceXAI moved it from API and developer access into the consumer Grok experience across web, X, iOS, and Android. That makes the model easier to try, although the vendor's performance claims still need to survive real workloads.

Grok 4.5 release timeline and availability

The separate dates explain why searches surface July 8, July 16, July 22, and July 23. The official model announcement is dated July 16, while SpaceXAI's consumer rollout page is dated July 22.

The week's other verified AI changes

• GPT-5.6 moved into general availability on July 9. OpenAI launched Sol, Terra, and Luna across ChatGPT, Codex, and the API, with different capability and price tiers.
• NotebookLM became Gemini Notebook on July 16. Google added a secure cloud-computer workflow that can write and execute code while staying grounded in notebook sources.
• Notion Agents gained calendar tools on July 16. Agents can show and manage schedules, join calls, send invites, and create scheduling links from chat on desktop.

The headline change: Grok 4.5 arrives with frontier-model pricing pressure

SpaceXAI announced Grok 4.5 on July 16 for coding, agentic tasks, and knowledge work. The model was available through the SpaceXAI API and console, Grok Build, and Cursor. On July 22, SpaceXAI expanded it across grok.com, X, iOS, and Android, adding the conversational availability reflected in many current searches. Grok Build now uses it as the default model, and SpaceXAI is offering free use in Grok Build and Cursor for a limited time.

The number that matters immediately is the API price: $2 per million input tokens and $6 per million output tokens. That puts Grok 4.5 in a price band normally associated with smaller, faster models rather than a vendor's flagship agentic model.

SpaceXAI also says Grok 4.5 is served at 80 tokens per second and uses fewer output tokens than selected competitors on its published engineering evaluations. Those are vendor claims, not independent proof, so they should be treated as a reason to test rather than a reason to migrate.

Choosely's read: Grok 4.5 deserves a controlled trial for teams running repeated coding, research, document, spreadsheet, or agent tasks where token costs compound quickly. It does not deserve a blind replacement decision. The right comparison is cost per successful task, including retries, review time, tool-call reliability, and output quality, not price per token alone.

Who should care

• Developers and agent builders with high-volume API workloads.
• Cursor users who can test the model without changing their editor.
• Teams using Grok Build for code, Excel, PowerPoint, or Word workflows.
• Buyers currently paying premium-model rates for routine agent steps.

What to do next

Run the same representative workload through Grok 4.5 and your current default. Track successful completion rate, output tokens, latency, retries, and human correction time. If Grok wins only on token price but creates more review work, it is not actually cheaper.

Read SpaceXAI's official Grok 4.5 announcement. See the web, X, iOS, and Android rollout.

GPT-5.6 is now generally available across three tiers

OpenAI moved the GPT-5.6 family from limited preview to general availability on July 9. The family has three tiers: Sol as the flagship, Terra as the balanced everyday model, and Luna as the lowest-cost option.

GPT-5.6 is available across ChatGPT, Codex, and the OpenAI API. Access varies by product and plan. In ChatGPT Work and Codex, paid users can choose among the models and reasoning-effort settings; OpenAI also introduced higher-compute modes for demanding work. Through the API, all three models are available in the Responses API.

Published API pricing per million tokens is:

OpenAI reduced Terra and Luna pricing on July 30. Terra now costs $2 per million input tokens and $12 per million output tokens, while Luna costs $0.20 input and $1.20 output. Sol pricing is unchanged.

The tiering matters more than a single "best model" headline. Teams can reserve Sol for difficult work, use Terra as the default, and route simpler high-volume steps to Luna without leaving one provider's ecosystem.

Choosely's read: If you already use OpenAI, the immediate job is routing, not wholesale migration. Audit where you are overpaying for easy work and where higher reasoning effort actually changes the result. The family gives operators more control, but it also creates more configuration decisions.

Read OpenAI's GPT-5.6 launch announcement. See OpenAI's July 30 pricing update.

NotebookLM becomes Gemini Notebook and starts doing the work

Google renamed NotebookLM to Gemini Notebook on July 16. The product remains available as a standalone notebook, while also becoming more closely connected to the Gemini app.

The more material change is a secure cloud computer that can work from notebook sources, write and execute code, and complete multi-step tasks. Google says it is rolling out first to Google AI Ultra users and eligible Workspace customers, with broader access for Google AI Pro users on the web over the following weeks.

That moves the product beyond source-grounded summaries and audio overviews. A notebook can now become a working environment where the model analyses source material, creates code when useful, and produces an artifact or result without forcing the user to move the context into a separate tool.

Choosely's read: Research-heavy teams should test this on a bounded, verifiable job: analysing a set of reports, cleaning a dataset, or building a small evidence-backed model. Source grounding reduces one kind of risk, but executed code and multi-step automation introduce another. Review what data the notebook can access, what it produces, and whether the result can be reproduced.

Read Google's official Gemini Notebook announcement.

Notion Agents can now manage calendars from chat

Notion's July 16 release added calendar tools for agents. From a Notion chat, an agent can show a schedule, move a meeting, join calls, send invitations, find suitable times, and create scheduling links.

The feature is available on desktop, with mobile support still to come. Users can connect a calendar from the agent chat or through Notion's Connections settings.

This looks like a small product update, but it closes an important workflow gap. An agent that can read project context but cannot act on the calendar still leaves the final coordination step to a human. Calendar actions let a planning conversation turn directly into a changed schedule.

Choosely's read: Treat calendar access as operational permission, not a convenience toggle. Start with a low-risk test calendar or narrowly scoped use case. Confirm how the agent handles time zones, invitees, conflicts, and destructive changes before relying on it for external meetings.

Read Notion's official July 16 release notes.

The pattern behind this week's changes

The four announcements point in the same direction:

1. 1Models are being packaged as workers. Grok 4.5 and GPT-5.6 are being sold around coding, tools, and sustained professional tasks rather than chat quality alone.
2. 2The work surface is moving closer to the context. Gemini Notebook works inside source collections; Notion Agents act inside calendars and workspace context.
3. 3Routing is becoming a core operator skill. More model tiers and sharper prices are useful only when teams know which work belongs on which model.
4. 4Permissions deserve the same attention as prompts. A model that can execute code or change a calendar can save time, but it can also make consequential mistakes faster.

The sensible response is not to subscribe to everything. Pick one workflow where the new capability could remove measurable effort, define a success test, and keep the old path available until the new one proves reliable.

What Choosely verified this week

Every material launch date, availability statement, and published price in this brief was checked against a first-party vendor announcement or release page, with GPT-5.6 pricing reverified on July 31, 2026.

• SpaceXAI: Introducing Grok 4.5
• SpaceXAI: Bringing Grok 4.5 to iOS, Android, Web, and X
• SpaceXAI Docs: Release Notes
• OpenAI: GPT-5.6
• OpenAI: July 30 GPT-5.6 pricing update
• Google: NotebookLM becomes Gemini Notebook
• Notion: What's New

Availability can still vary by plan, geography, workspace settings, and staged rollout. Check the relevant account before changing a production workflow.

The bottom line

Grok 4.5 is the week's clearest pricing signal: a flagship agentic model at $2 input and $6 output per million tokens, now available through developer tools and the consumer Grok experience. GPT-5.6 answers with a three-tier family, while Google and Notion are embedding agents directly into research and scheduling workflows.

The winning tool is not automatically the newest or cheapest. It is the one that completes your real task reliably, at a total cost you can defend, with permissions you understand.