Grok API Pricing
Elon Musk's AI lab, founded March 2023 as X.AI Corp. Ships the Grok family — frontier reasoning and coding models trained on the Memphis Colossus supercluster — 200,000 H100s in one interconnected cluster, 150 MW at peak, with a stated roadmap to 1M GPUs. Acquired by SpaceX on Feb 2, 2026 at a $250B valuation ($1.25T combined entity). Renamed from xAI to SpaceXAI in July 2026, when the lab folded fully into SpaceX. The Grok brand is unchanged, the API still lives at docs.x.ai, and Grok stays integrated natively into X (Twitter).
Grok 4.7 LAUNCHED SEP 2026
SpaceXAI's newest frontier model for coding, agentic tasks and knowledge work: a new, larger base model than Grok 4.6, sold on the same rate card.
Grok 4.6 LAUNCHED AUG 2026
SpaceXAI's previous frontier model for coding, agentic tasks and knowledge work, succeeded on the same rate card by Grok 4.7 on 2026-09-21.
Grok 4.5 LAUNCHED JUL 2026
xAI pricing docs list grok-4.5 at $2.00/M input, $0.30/M cached input, and $6.00/M output below 200K prompt tokens.
Grok 4.3 LAUNCHED MAY 2026
The cheaper 1M-context Grok, still listed alongside the 500K-context line that starts at Grok 4.5.
Grok 4.20 (0309) Reasoning
Dated 03-09 snapshot of Grok 4.20 in reasoning mode.
Grok 4.20 (0309) Non-Reasoning
Dated 03-09 snapshot of Grok 4.20 in non-reasoning (latency-optimised) mode.
Grok 4.20 Multi-Agent (0309)
Multi-agent orchestration variant of Grok 4.20; xAI describes it as "Multiple agents collaborate in parallel to perform deep research tasks." Same per-token price as grok-4.3 ($1.25/M input, $2.50/…
Grok Build 0.1 LAUNCHED MAY 2026
xAI's fast agentic-coding model, now listed on the standard Code API pricing table at $1/M input and $2/M output (256K context).
Grok 4.1 Fast LAUNCHED NOV 2025
Archive price retained from the verified snapshot: $0.20/M input, $0.05/M cached input, $0.50/M output.
Grok 4 Fast LAUNCHED SEP 2025
Archive price retained from the verified snapshot: $0.20/M input, $0.05/M cached input, $0.50/M output.
Grok 4 LAUNCHED JUL 2025
Archive price retained from the verified snapshot: $3/M input, $0.75/M cached input, $15/M output.
Grok 3 LAUNCHED FEB 2025
Archive price retained from the verified snapshot: $3/M input, $0.75/M cached input, $15/M output.
Grok 3 Mini LAUNCHED FEB 2025
Archive price retained from the verified snapshot: $0.30/M input and $0.50/M output.
Grok 2 (1212) LAUNCHED DEC 2024
Official xAI 1212 launch post announced grok-2-1212 and grok-2-vision-1212 at $2/M input and $10/M output after efficiency improvements.
Grok 2 Vision (1212) LAUNCHED DEC 2024
API ID grok-2-vision-1212.
Grok Code Fast 1 LAUNCHED AUG 2025
Archive price retained from the verified snapshot: $0.20/M input, $0.02/M cached input, $1.50/M output.
Grok Beta LAUNCHED NOV 2024
API ID grok-beta.
Grok Vision Beta LAUNCHED NOV 2024
API ID grok-vision-beta.
| Model | Input /M | Output /M | Cached | Context | Max output | Vision | Tools | Tier |
|---|---|---|---|---|---|---|---|---|
| Grok 4.7 FLAGSHIP | $2.00 | $6.00 | $0.50−75% | 500K | — | ✓ | ✓ | Frontier |
| Grok 4.6 | $2.00 | $6.00 | $0.50−75% | 500K | — | ✓ | ✓ | Frontier |
| Grok 4.5 | $2.00 | $6.00 | $0.30−85% | 500K | — | ✓ | ✓ | Frontier |
| Grok 4.3 | $1.25 | $2.50 | $0.20−84% | 1M | — | ✓ | ✓ | Frontier |
| Grok 4.20 (0309) Reasoning | $1.25 | $2.50 | $0.20−84% | 1M | — | ✓ | ✓ | Frontier |
| Grok 4.20 (0309) Non-Reasoning | $1.25 | $2.50 | $0.20−84% | 1M | — | ✓ | ✓ | Frontier |
| Grok 4.20 Multi-Agent (0309) | $1.25 | $2.50 | $0.20−84% | 1M | — | ✓ | ✓ | Frontier |
| Grok Build 0.1 | $1.00 | $2.00 | $0.20−80% | 256K | — | ✓ | ✓ | Active |
| Grok 4.1 Fast RETIRED | — | — | — | Retired May 15, 2026 | → Grok 4.3 | ✗ | ✗ | Retired |
| Grok 4 Fast RETIRED | — | — | — | Retired May 15, 2026 | → Grok 4.3 | ✗ | ✗ | Retired |
| Grok 4 RETIRED | — | — | — | Retired May 15, 2026 | → Grok 4.3 | ✗ | ✗ | Retired |
| Grok 3 RETIRED | — | — | — | Retired May 15, 2026 | → Grok 4.3 | ✗ | ✗ | Retired |
| Grok 3 Mini RETIRED | — | — | — | Retired Sep 30, 2025 | → Grok 4.3 | ✗ | ✗ | Retired |
| Grok 2 (1212) RETIRED | — | — | — | Retired Aug 14, 2025 | → Grok 4.3 | ✗ | ✗ | Retired |
| Grok 2 Vision (1212) RETIRED | — | — | — | Retired Aug 14, 2025 | → Grok 4.3 | ✓ | ✗ | Retired |
| Grok Code Fast 1 RETIRED | — | — | — | Retired May 15, 2026 | → Grok 4.3 | ✗ | ✗ | Retired |
| Grok Beta RETIRED | — | — | — | Retired Feb 28, 2025 | → Grok 4.3 | ✗ | ✗ | Retired |
| Grok Vision Beta RETIRED | — | — | — | Retired Feb 28, 2025 | → Grok 4.3 | ✓ | ✗ | Retired |
Pricing across the lineup.
The last 12 months at SpaceXAI.
Major releases, retirements, and corporate moves. Sourced from SpaceXAI press releases, docs.x.ai migration guides, and recent merger filings.
$2/$6 card; it becomes Grok Build's default model, and a Fast variant at twice the rates is sold only in Cursor and Grok Build$2/$6 with a $0.50 cache read and reasoning effort up to xhigh; the next day Grok 4.5's cached input was cut to $0.30$2/$6, 500K context, configurable reasoning$1/$2console.x.ai + docs.x.ai — the direct API, OpenAI-compatible Chat Completions plus a Responses API. A US regional endpoint, us.api.x.ai, keeps inference in the United States at 1.1x the rates for Grok 4.7 and 4.6. SpaceXAI's docs also list OpenRouter, Vercel and Cloudflare as gateways serving Grok 4.7.
grok.com for standalone chat, X (Twitter) for in-feed integration. SuperGrok at $30/mo, SuperGrok Heavy at $300/mo (priority access to the largest variant).
AGENT · HOSTEDSpaceXAI's work agent — persistent bots on a hosted cloud computer, not a coding tool. Sold with no price of its own: bundled into SuperGrok and above, and into every paid Cursor plan, where the account and the meter actually live.
AGENT · TERMINALSpaceXAI's coding agent — the one the name Grok Bot is mistaken for. Runs locally in a project directory, harness and TUI open source under Apache-2.0, and included on every Grok plan including the free one. Its allowance is not its own: one weekly pool covers Chat, Imagine, Voice, the API and Build together.
CLOUD · RESELLERSSpaceXAI announced Grok 4.6 on Amazon Bedrock (Aug 19, 2026), Gemini Enterprise Agent Platform (Aug 21), Microsoft Foundry (Aug 26) and GitHub Copilot (Aug 14); the docs add OpenRouter, Vercel and Cloudflare as gateways for Grok 4.7. Each platform bills at its own rates, which are not the vendor rates on this page.
INFRA · COMPUTESpaceXAI's own supercluster in Memphis, TN. Built in 122 days, then doubled in 92 more to 200,000 H100s in a single interconnected cluster, drawing 150 MW at peak, with a stated roadmap to 1M GPUs. Training and inference capacity, not a customer-facing endpoint.
SpaceXAI was founded by Elon Musk on March 9, 2023 as X.AI Corp, with 11 founding researchers (including ex-DeepMind's Igor Babuschkin). The pitch: build a maximally truth-seeking AI inside Musk's ecosystem — X (Twitter) for data and distribution, Tesla / SpaceX for compute and capital. CEO: Elon Musk.
The lab moved fast. Grok 1 shipped November 2023 (weights later open-sourced). Grok 2 in August 2024 added image generation. Grok 3 in February 2025 introduced the reflection / reasoning mode that anchors the current line. Grok 4 in July 2025 was the first frontier-tier Grok ($3/$15) and shipped alongside SuperGrok Heavy, a $300/mo consumer tier that exposed early access to the largest variants. Grok 4.3 on May 12, 2026 cut input prices ~40% to $1.25/$2.50 and the next-generation Grok 4.20 family arrived in 03-09 snapshot form.
The defining infrastructure asset is the Memphis Colossus supercluster, brought online in December 2024 after a remarkable 122-day build. Peak draw 150 MW; expansion plans target 1M+ GPUs.
The corporate structure changed dramatically in 2026. On Feb 2, 2026, SpaceX acquired SpaceXAI at a $250B SpaceXAI valuation and $1.25T combined-entity valuation — the largest tech merger by valuation on record. Stated rationale: orbital data centres + more capital runway for AI buildout against OpenAI and Anthropic. SpaceXAI continues to operate under its own brand and API surface; only the parent changed.
Compared to Anthropic and OpenAI: cheaper per token at the frontier (Grok 4.7 at $2.00/$6.00 per 1M against Claude Opus 5 at $5.00/$25.00), a narrower surface (the Grok text models take text and image input and return text, while image generation, video and voice live in the separate Imagine and Voice APIs), and tied to the X / Twitter brand for better and worse.
OpenAI
Bigger surface area (image / audio / video / embeddings under one roof), bigger consumer brand. GPT-6 Astra, OpenAI's top model, lists at $10.00 input per 1M against Grok 4.7's $2.00. SpaceXAI wins on per-token cost; OpenAI wins on multimodality and ecosystem.
CODING LEADERAnthropic
The coding-agent incumbent. Claude Opus 5 lists at $5.00 input and $25.00 output per 1M against Grok 4.7's $2.00 and $6.00, and Anthropic publishes a Responsible Scaling Policy. On per-token price alone, Grok is the cheaper bet.
CLOUD GIANTAudio and video in, which Grok has nowhere. Gemini 3.8 Flash takes text, images, audio and video, keeps a 1M window, and at $0.75/$3.75 per 1M it undercuts Grok 4.7 — price is not SpaceXAI's edge at this tier. Distribution through Workspace and Vertex is the other gap.
Meta
Muse Glimmer is the self-hostable one — 30B parameters, Apache 2.0, and Meta says it runs on a single GPU. Its frontier Muse Spark line is API-only, so the trade is the mirror of SpaceXAI's: Meta gives you weights at the small end, SpaceXAI gives you none (Grok 1 was released in 2024, nothing since).
Alibaba (Qwen)
The Qwen (Tongyi Qianwen) model family from Alibaba Cloud's Tongyi Lab. First released in 2023, Qwen is the most-downloaded open-weight model family in the world — most tiers ship under Apache 2.0 on Hugging Face and ModelScope, while the proprietary Max tier is served through Alibaba Cloud's Model Studio.
GLM ARCHITECTZhipu (Z.ai / GLM)
The Beijing lab behind the GLM models, also operating internationally as Z.ai. Spun out of Tsinghua University in 2019, Zhipu ships strong coding/agentic models, multiple free tiers, and open weights — and was the first of China's "Six Tigers" to pursue an IPO. It also sits on the US Entity List.
Frequently asked.
Practical questions about SpaceXAI, the Grok lineup, the SpaceX merger, and how text-only frontier pricing actually shakes out.
Q · 01 Which Grok model should I start with? +
Q · 02 Does Grok support vision, audio, or image generation? +
Q · 03 Did SpaceX really acquire SpaceXAI? +
Feb 2, 2026 at a $250B SpaceXAI valuation ($1.25T combined entity). The deal was framed as funding for orbital data centres and giving SpaceXAI more capital runway against OpenAI / Anthropic infrastructure spend. The SpaceXAI brand, console at console.x.ai, and Grok product line all continue to operate as before; only the parent ownership changed.Q · 04 Does SpaceXAI offer EU data residency? +
us-east-1, us-west-2, us-central-1), and the one regional option on the pricing page is a US endpoint. Grok 4.5 did become available in the API console for EU users in July 2026, but that is availability, not EU processing. Since August 2026 Grok 4.6 is also sold through Amazon Bedrock, Microsoft Foundry and the Gemini Enterprise Agent Platform; whether any of them offers Grok in an EU region is that platform's answer to give, and we have not verified it. If EU data residency is a hard requirement, pick Claude on Bedrock EU, Gemini on Vertex EU, or Mistral.Q · 05 What happened in the May 15, 2026 retirement wave? +
grok-4.3 and bill at Grok 4.3 pricing ($1.25/$2.50). That left four active text models at the time; Grok Build 0.1, Grok 4.5, Grok 4.6 and Grok 4.7 have been added since. Migration guide: docs.x.ai/developers/migration/may-15-retirement.