Grok 4.7: Benchmarks, Pricing, vs Claude, and Limits (2026)

Grok 4.7 is xAI's newest frontier reasoning model, built with native multi-step thinking and a 1-million-token context window. While xAI prices its raw API under Claude 5.5 on headline output tokens, independent testing shows Grok 4.7 produces more than double the output tokens per task, narrowing any cost advantage. For developers and technical operators who want to test Grok 4.7 alongside Claude and Gemini without paying for a separate SuperGrok tier, multi-model platforms let you run it on an existing plan today.

Why Grok 4.7 matters right now

xAI rolled out Grok 4.7 to challenge Anthropic's Claude 5.5 family and OpenAI's GPT-6 lineup in terminal execution, agentic coding, and extended reasoning. Unlike standard conversational models where reasoning is an optional toggle, Grok 4.7 operates as an always-on reasoning model. It explores intermediate thinking paths before returning its final answer.

Developers are paying close attention because xAI advertised significant gains on terminal coding benchmarks. However, the first round of independent evaluations reveals important tradeoffs between raw model capability, token efficiency, and response latency.

Independent benchmarks: reality vs marketing claims

Marketing releases often report benchmark numbers tested under custom evaluation harnesses. Third-party testing from Artificial Analysis, measured under rigorous independent conditions as of November 2026, provides a clearer picture of how Grok 4.7 actually performs:

Three findings stand out from the data:

First, on Terminal-Bench 4.0, xAI reported a 37.6% success rate on its internal harness. When tested independently by Artificial Analysis at high reasoning effort, Grok 4.7 scored 25.8%. That represents a genuine 4.5 percentage point improvement over Grok 4.6, but it leaves Grok substantially behind Claude Sonnet 5.5 (54.2%) and Claude Opus 5.5 (58.5%).

Second, Grok 4.7 performs well on structured knowledge synthesis. Its AA-Briefcase score of 1,657 Elo edges past Claude Sonnet 5.5 (1,642 Elo), demonstrating solid competence in digesting business documents, drafting structured briefs, and reconciling contradictory facts.

Third, latency is noticeable. Because Grok 4.7 reasons deeply across broad trajectory trees, it averaged nearly 11 minutes per complex task on benchmark evaluations, compared to roughly 3 minutes for Claude Opus 5.5 and under 3 minutes for Claude Sonnet 5.5.

Pricing and the hidden cost of verbosity

On paper, xAI's direct API rates look competitive as of November 2026:

• Prompts under 200,000 tokens: $2.00 per million input tokens, $6.00 per million output tokens.
• Prompts of 200,000 tokens or more: prices double to $4.00 per million input tokens and $12.00 per million output tokens.
• Prompt cache reads: $0.50 per million tokens.

By comparison, Anthropic charges $2.00 input and $10.00 output per million tokens for Claude Sonnet 5.5, with prompt cache reads at $0.20 per million tokens (a 90% caching discount). Claude Opus 5.5 sits at $4.00 input and $20.00 output per million tokens.

However, token pricing alone does not determine your final bill. Independent benchmark tracking shows that Grok 4.7 generates roughly 2.25 times as many output tokens as Claude 5.5 to solve equivalent programming problems. Because its reasoning tokens count toward billable output, a task requiring 1,200 output tokens on Sonnet 5.5 often consumes 2,700 output tokens on Grok 4.7.

When you factor in Grok's $0.50 cache read price compared to Anthropic's $0.20 rate, multi-turn agent loops that repeatedly inspect context can end up costing more on Grok 4.7 than on Claude Sonnet 5.5.

How to run Grok 4.7 in Zemith

If you want to use Grok 4.7 without paying $30 to $40 every month for an isolated xAI web subscription or managing per-token API billing, you can run it directly inside Zemith.

Grok 4.7 is included on the Zemith Plus plan ($15.99 per month) and Professional plan ($29.99 per month). Here is the workflow:

1. Open Chat in Zemith on web, iOS, or Android.
2. Click Select model at the top of the conversation.
3. Choose Grok 4.7 under the xAI provider list.
4. To analyze long documents or codebases, add your files to a Library Project (Plus supports up to 50 source documents) and ground Grok 4.7 in those files.
5. If Grok encounters an edge case where it gets stuck in an extended reasoning loop, switch models mid-chat to Claude 5.5 Sonnet or Gemini 3.8 Flash without losing conversation history.

In Zemith, Grok 4.7 costs 48 credits per 100 input tokens, 144 credits per 100 output tokens, and 12 credits per 100 cached input tokens. A typical complex reasoning request consuming 2,000 input tokens and 800 output tokens uses approximately 2,112 credits. On a Plus plan with 1.5 million monthly credits, that provides hundreds of deep reasoning sessions each month alongside your regular access to Claude, GPT, and Gemini.

Is Grok 4.7 better than Grok 4.6?

Yes, in measurable ways. Grok 4.7 is a clear upgrade over Grok 4.6 across reasoning depth, agentic workflow handling, and coding accuracy.

Artificial Analysis measured Grok 4.7's Intelligence Index at 46 (up from 44 on Grok 4.6) and its Coding Agent Index at 56 (a 9-point jump over Grok 4.6's 47). Its performance on real-world business scenarios (AA-Briefcase) climbed by 111 Elo points to 1,657.

The primary disadvantage compared to Grok 4.6 is response time. Because Grok 4.7 engages deep reasoning pathways by default, routine questions that Grok 4.6 answered in seconds now take longer while the model generates its internal thought chain.

Grok 4.7 vs Claude Opus 5.5 and Sonnet 5.5: which should you pick?

The choice between Grok 4.7 and Anthropic's Claude 5.5 models comes down to task type:

• Pick Claude 5.5 Sonnet for daily software engineering, fast refactoring, and multi-turn iterative agent workflows. Sonnet 5.5 delivers 85 to 139 tokens per second, scores nearly 30 percentage points higher on Terminal-Bench 4.0, and benefits from an aggressive 90% prompt caching discount that keeps long chat sessions inexpensive.
• Pick Claude 5.5 Opus when you need maximum architectural judgment, mission-critical correctness on complex codebases, or nuanced analytical writing where mistakes are costly.
• Pick Grok 4.7 for knowledge synthesis across dense reports, exploratory research where you want broad reasoning trajectories, and single-pass queries that benefit from deep deliberation without needing rapid turnarounds.

Because both models excel at different tasks, locking yourself into a single vendor's subscription forces you into compromises. Running both in a unified workspace gives you the strongest tool for each job.

What does Grok 4.7 cost, and is SuperGrok worth it?

xAI offers Grok 4.7 through two consumer channels:

1. X Premium+ / SuperGrok: Pricing ranges from $16 to $40 per month depending on billing frequency and if purchased through an X platform add-on or a standalone subscription. However, users frequently hit 5-hour rolling session limits and queue throttling during heavy usage spikes.
2. xAI API: $2.00 input and $6.00 output per million tokens for prompts under 200,000 tokens, doubling above that threshold.

For most individual developers and professionals, paying $30 or more solely for SuperGrok is difficult to justify when you still need a separate subscription for Claude or ChatGPT. A multi-model subscription like Zemith Plus ($15.99 per month) gives you Grok 4.7, Claude 5.5 Sonnet, GPT 6.1 Terra, and Gemini 3.8 Flash in one place for less than the price of a single standalone subscription.

To see how combined plans compare with individual subscriptions, see our guide to getting ChatGPT Plus and Claude Pro in one subscription.

Can Grok 4.7 run code and search the live web?

In Zemith, Grok 4.7 has tool access enabled. When you ask a question that requires current information, the model can execute web searches with cited sources. It can also analyze files and execute scripts through sandboxed computation.

Note that Grok 4.7 does not support direct file drag-and-drop in an ungrounded chat window (fileUpload: false). To work with documents, spreadsheets, or code repositories, you create a Library Project, attach your source files there, and select Grok 4.7 as your project chat model.

Honest caveats and what to keep in mind

Before switching your primary workflows to Grok 4.7, keep these limitations in mind:

• No reasoning toggle: Grok 4.7 always reasons. You cannot turn off the thinking phase to get quick, low-latency answers for simple questions. For quick edits or trivial lookups, Claude 5.5 Haiku or Gemini 3.8 Flash will feel much faster.
• Coding gap on terminal tasks: Despite vendor marketing emphasizing coding gains, independent benchmarks show Grok 4.7 (25.8%) still trails frontier coding models like Claude Opus 5.5 (58.5%) and GPT-6 Astra on complex software engineering benchmarks.
• Context price cliff: Once your prompt exceeds 200,000 tokens, xAI's API doubles its input and output rates. If you analyze massive codebases or video inputs, monitoring context size is critical.

For detailed comparisons of frontier models across benchmarks and coding limits, explore our breakdown of Gemini 4 Argon vs Claude Opus 5.5 and our overview of multi-model AI chat.

The bottom line

Grok 4.7 is a capable, highly deliberative reasoning model that represents a major step forward for xAI, especially in knowledge synthesis and exploratory research. But its high verbosity, longer response latency, and lower terminal coding scores mean it does not unseat Claude 5.5 as the default choice for software development.

Rather than committing to an expensive standalone SuperGrok plan, test Grok 4.7 in Zemith alongside Claude and GPT. You get the strengths of xAI's reasoning model when you need deep deliberation, with the freedom to switch to faster frontier models whenever speed and precision matter.