xAI Grok 4.7 — API Pricing Impact (Sep 2026)

xAI released Grok 4.7 on September 21 as its new frontier model for coding, agentic tasks, and knowledge work. The model is live as grok-4.7 on the public xAI API and is also available through Grok Build and Cursor. xAI names OpenRouter, Vercel, and Cloudflare as gateway partners, although the public OpenRouter model catalog did not enumerate the exact Grok 4.7 route at our launch check.

The headline pricing result is simple: Grok 4.7 keeps Grok 4.6’s token rates at both context tiers. The live chart, table, and calculator above read those rates from our updated canonical dataset instead of freezing them in this article.

Grok 4.7 pricing vs Grok 4.6

Grok 4.7 and 4.6 have identical standard input, cached-input, and output rates. They also share a 500K context window and the same higher price tier once a prompt reaches 200K tokens. The higher tier applies to all tokens in that request, so crossing the boundary can matter more than the version change.

Route	Price change vs Grok 4.6	Availability	Best use
Grok 4.7 public API	None	xAI API and gateways	Direct apps and agent systems
Grok 4.7 Fast	Separate premium	Cursor and Grok Build only	Latency-sensitive coding work
US regional endpoint	Regional premium	xAI US endpoint	Workloads requiring US inference
Priority processing	Processing premium	Supported synchronous API requests	Requests where lower latency has measurable value

xAI’s Fast documentation needs care. It describes Fast as twice the standard token rates, but its published table is two times the base tier and one-and-a-half times the public long-context tier. Buyers above the threshold should use the explicit Fast table and verify an invoice rather than applying a blanket multiplier.

What improved in Grok 4.7

xAI says 4.7 uses a larger base model than 4.6, a longer reinforcement-learning run, and harder training tasks weighted toward work that takes hours. It also adds native understanding of the Grok Bot harness and stronger self-verification.

xAI-reported benchmark	Grok 4.7	Grok 4.6	Change
CursorBench 4.0	46.3%	40.4%	+5.9 points
DeepSWE v1.1	71.0%	65.2%	+5.8 points
Terminal-Bench 4.0	38.0%	20.3%	+17.7 points
Harvey Legal Agent	19.6%	15.8%	+3.8 points
HealthBench Professional	56.7%	48.5%	+8.2 points
EEBench	64.0%	53.0%	+11.0 points

These are xAI’s evaluations, not independent AI Pricing Guru results. The launch post does not provide a complete reproduction package for every harness, and model effort settings differ across parts of the table. Treat the scores as reasons to test, not guaranteed production gains.

Features and migration details

Grok 4.7 accepts text and images and returns text with no stated output limit. It supports low, medium, high, and xhigh reasoning effort, with high as the default, and works through both Responses and Chat Completions APIs. The knowledge cutoff is May 2026.

On the Responses API, encrypted reasoning is always returned. xAI tells developers to pass those reasoning items back unchanged during multi-turn work. It also recommends prompt_cache_key—or the conversation ID header for Chat Completions—to improve cache affinity, plus context compaction for long agent loops.

Who benefits—and who should wait

Existing Grok 4.6 users have the clearest upgrade case because the list rate does not increase. Coding agents, terminal automation, document creation, and long-horizon tool workflows are the main targets of xAI’s reported gains.

Teams should wait for their own replay results when workflows are already stable, regulated, or highly sensitive to output length. Better reasoning can still produce a larger bill if it generates more tokens, calls more tools, or runs longer loops. Compare the updated xAI pricing page with OpenAI pricing and Anthropic pricing, then model your traffic in the token calculator.

For a fixed-price coding alternative, compare the GLM Coding Plan with measured Grok 4.7 usage.

Affiliate disclosure: AI Pricing Guru may earn a commission from the sponsored link above at no extra cost to you. It does not affect this analysis.

What developers should do now

1. Pin grok-4.7 in a canary instead of moving a shared alias globally.
2. Replay the same Grok 4.6 tasks at matched reasoning effort and output limits.
3. Test prompts on both sides of the long-context threshold.
4. Record cache hits, reasoning and output tokens, tool calls, retries, latency, and accepted-task rate.
5. Keep Grok 4.6 available for rollback until output schemas and tool behavior are stable.

Do not copy xAI’s benchmark scores into AI Pricing Guru Labs. Grok 4.6 already has a measured 49/49 result, but Grok 4.7 needs a fresh run through the same fixed suite with complete responses and spend logs. The maintained OpenRouter transport did not list the exact 4.7 route at our September 21 check, so the coverage note records a route blocker instead of publishing a proxy or relabeling the 4.6 result.

Bottom line

Grok 4.7 is a same-rate flagship upgrade with credible reasons for coding and agent teams to test immediately. It is not a forced migration and not proof of lower real-world cost. Canary it against 4.6, keep context below the higher tier where practical, and promote it only when completed-work economics improve.

Sources: xAI’s official Grok 4.7 announcement, release note, model overview, and API pricing. Availability, rates, features, and benchmark claims checked September 21, 2026.