Grok API Pricing 2026: Every xAI Model, Compared and Priced Against Rivals

xAI’s price list has an unusual shape: three flagship generations — Grok 4.5, 4.6 and 4.7 — all cost exactly $2 input / $6 output per 1M tokens, and the model below them, Grok 4.3, costs about half as much at a typical mix while offering double the context. So the Grok pricing question is not really “how much does Grok cost?” It is “why would you pay the flagship rate at all, and when does the 200K long-context rule quietly double your bill?” This page answers both, then prices Grok against the rival flagships at one standard workload.

Prices are xAI’s official API list prices from docs.x.ai, verified 2026-10-28, the same source behind our xAI provider page. Standard workload for monthly figures: 10M tokens, split 8M input / 2M output, prompts below 200K tokens, no caching, standard tier. Grok subscriptions (SuperGrok, X Premium) are a separate product and are not mixed in; Imagine image/video and Voice APIs have per-image and per-minute schedules and are outside this token comparison.

Disclosure: this article contains no affiliate links. Where CostMyTokens does use affiliate links elsewhere, they never affect order — every table is sorted by price only. Read our disclosure.

The xAI price list

Four more models sit on the official list without model pages here yet. The three Grok 4.20 variants (reasoning, non-reasoning, multi-agent) price identically to Grok 4.3 — $1.25 / $2.50, cached $0.20, 1M context. Grok Build 0.1, xAI’s coding model, is the cheapest row on the list at $1.00 / $2.00 (cached $0.20) with a 256K context — $12.00 at the standard workload. Every model above also has a long-context schedule, covered next — it is the most important line on the page.

Three flagships, one price

At $2 / $6, Grok 4.5, 4.6 and 4.7 are indistinguishable on the standard columns. The only price difference inside the trio is cached input: Grok 4.5 reads cache at $0.30 per 1M against $0.50 for 4.6 and 4.7 — a small edge that matters only on cache-heavy traffic (at 50% of input cached, 4.5 costs $21.20 per 10M against $22.00 for 4.7). On everything else, a newer flagship at the same price makes the older ones a pinning decision, not a saving: there is no discount for choosing 4.5 or 4.6 over Grok 4.7.

Grok 4.3 is where the actual price decision lives. At $1.25 / $2.50 it costs $15.00 at the standard workload — 46% less than the flagships — and its context window is 1M tokens, double the flagships’ 500K. For high-volume or long-document work that does not need the newest model, 4.3 is the rational default and the flagship is the upgrade to justify, not the other way round.

The 200K rule: one threshold, whole-request doubling

Every model on the list carries a long-context schedule that starts at a 200K-token prompt — and once a request crosses it, the long rates apply to all tokens in that request, not just the excess:

Cached input doubles too ($0.50 → $1.00 on Grok 4.7 and 4.6, $0.30 → $0.60 on 4.5). This is a harsher mechanic than OpenAI’s long-context segment, which starts above 272K tokens and doubles input and cached input but raises output only 1.5x; on xAI, output doubles as well, and the threshold is lower. An agent that accumulates history will cross 200K mid-conversation and stay there — budget agentic workloads at the long rate from the start, or trim context deliberately.

Grok against the rivals

Same standard workload (10M tokens, 80/20, short-context rates, no caching), official list prices, sorted by cost:

Grok 4.3 also undercuts OpenAI’s nearest tier, GPT-6.1 Terra, despite Terra’s lower $1 input — its $2.50 output is half of Terra’s $5. The pattern to notice at the top: Grok’s flagship matches the mainstream rivals on input ($2) and undercuts them 40% on output ($6 vs $10). Output is where chat and agent bills accumulate, so at real-world mixes Grok 4.7 lands between the value tier and the mainstream tier — cheaper than GPT-6.1 Sol and Sonnet 5.5, dearer than DeepSeek’s flagship, and nowhere near Astra or chat-latest money. This table ranks price only; it says nothing about which model answers best, and xAI’s own consumer differentiator — live X and web data — is not a token feature at all. It is a tool, priced in the next section.

Where Grok gives ground back is caching: its cached-input discount is 75% on the flagships ($0.50 vs $2.00), against 95% on GPT-6.1 Sol ($0.10). Cache-heavy workloads narrow Grok’s lead — at 50% of input cached, Grok 4.7 costs $22.00 per 10M against Sol’s $28.40, so Grok still wins at this mix, but a workload that is almost entirely cached prompt reads its input at five times Sol’s rate.

Discounts and multipliers

• Batch: 20% off, on every text model above. xAI’s Batch API discount is 20% for Grok 4.7, 4.6, 4.5, 4.3 and the 4.20 variants (not Build 0.1), applied to all token types. That is a shallower cut than OpenAI’s 50% Batch or DeepSeek’s 50% off-peak — plan on 20%, not half. Batch Grok 4.7 costs $22.40 at the standard workload; Batch Grok 4.3 costs $12.00.
• Server-side tools are per-call. Web Search and Code Execution cost $5 per 1,000 calls; X Search costs $5 per 1,000 posts ($10 per 1,000 profiles); Collections Search $2.50 per 1,000 calls. An agent that searches X on every request adds a real second line to the token bill.
• Priority Processing: 2x. Priority requests bill at twice the standard rates across all token types.
• US regional endpoint: 1.1x. Requests on the US endpoint (currently Grok 4.7 and 4.6 only) carry a 10% premium — $30.80 at the standard workload on 4.7.
• Grok 4.7 Fast is not an API tier. The Fast variant (2x standard rates below 200K prompt tokens, 1.5x on long-context requests) is sold only inside Cursor and Grok Build and billed through those plans; it never appears on the public API price list.

No free credits to budget around

xAI’s earlier API credit programs — the signup credit and the monthly data-sharing credit — have ended, and the API is now straightforward prepaid / pay-as-you-go at the rates above. How xAI compares on free access across providers is covered in free LLM API tiers compared.

Verdict

• Default to Grok 4.3 for volume, long documents and anything cost-sensitive: $15.00 at the standard workload, 1M context, same 20% Batch discount.
• Pay the flagship rate for Grok 4.7 when you need the current model — at $28.00 it is still cheaper than the mainstream GPT and Claude tiers at the same mix, thanks to the $6 output rate.
• Skip 4.5 and 4.6 unless you are pinned to them: same price as 4.7, no saving, and only 4.5’s cheaper cached input ($0.30) can justify it on cache-heavy work.
• Watch the 200K line on any workload with growing context — it doubles the whole request, and it is the single most likely cause of a Grok bill arriving at twice the estimate.

Run your own mix on the CostMyTokens calculator, or price each model on its page: Grok 4.3, Grok 4.5, Grok 4.6 and Grok 4.7. The full catalog is on our xAI provider page; for the neighbouring price lists see ChatGPT API pricing and DeepSeek V4 Flash vs V4 Pro.

Prices: xAI API pricing, re-checked 2026-10-28; rival rows from each maker’s official list as tracked in our dataset. Prices change; check the official page before committing budget.