grok-4.6Grok 4.6 is xAI's frontier model, built around long-running agents and multi-step work — sustaining a task across many turns of research, codebase edits, or building an artifact. It matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index (61) and leads its predecessor on every published eval. 500K context, text and image input, and four reasoning-effort levels (low / medium / high — the default — / xhigh). Priced at $1.765 input / $5.294 output per 1M tokens, 12% under xAI official ($2.00 / $6.00), with cached input at $0.441. Called through the OpenAI-compatible /v1/chat/completions endpoint; /v1/responses also works and is the one that runs live web search. Failed requests are never billed.
View complete API reference with all parameters and examples.
View complete API reference with streaming, thinking, and more.
Billing: Cost = (input_tokens * input_price + output_tokens * output_price) / 1,000,000
$1.765 input / $5.294 output per 1M against official $2.00 / $6.00, with cached input at $0.441. Same model, same weights — the price difference is not a lesser tier
The 4.6 release targets tasks that span many steps — researching a topic, working across a codebase, turning an idea into a finished artifact — with more self-checking along the way. 500K-token context
low / medium / high / xhigh via reasoning_effort. Measured on our channel: 464 / 1,720 / 2,663 / 3,341 reasoning tokens on the same prompt. Unlike GPT-5.6, xhigh really is the deepest tier here
12/12 correct across 12 concurrent calls using a nested schema with enums and arrays — arguments parsed cleanly with no cross-request mixing. Streaming, multi-turn and parallel tool calls all pass
Cached input bills at $0.441 instead of $1.765. We send a per-conversation routing key so repeat turns keep landing on a warm server — without it, cache hits drop back to cold on roughly every other repeat
The default reasoning_effort is high, which runs about 53 seconds — set a client timeout above 60s. And billed output includes reasoning tokens, so your bill can exceed the visible answer; lower the effort when you do not need the depth
Grok 4.6은(는) xAI의 대규모 언어 모델 API입니다. Grok 4.6 is xAI's frontier model, built around long-running agents and multi-step work — sustaining a task across many turns of research, codebase edits, or building an artifact. It matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index (61) and leads its predecessor on every published eval. 500K context, text and image input, and four reasoning-effort levels (low / medium / high — the default — / xhigh). Priced at $1.765 input / $5.294 output per 1M tokens, 12% under xAI official ($2.00 / $6.00), with cached input at $0.441. Called through the OpenAI-compatible /v1/chat/completions endpoint; /v1/responses also works and is the one that runs live web search. Failed requests are never billed. APIMODELS 플랫폼을 거치면 통합 API와 투명한 종량 과금으로 이 모델을 호출할 수 있습니다. 현재 가격: Input: $1.765, Output: $5.294 per 1M tokens.
문의에 자동으로 답하는 대화 시스템을 만들어 응대 효율을 높입니다.
기사와 이메일, 광고 카피 같은 글을 자동으로 써서 작업량을 줄입니다.
코드 작성과 디버깅, 리뷰를 도와 개발 속도를 올립니다.
비정형 데이터를 읽어내 핵심을 뽑고 요약 리포트로 정리합니다.
Grok 4.6은(는) APIMODELS를 통해 Input: $1.765, Output: $5.294 per 1M tokens에 이용할 수 있습니다. 과금은 종량제라서 생성한 만큼만 냅니다.
APIMODELS에 가입해 API 키를 받고 통합 엔드포인트를 호출하면 됩니다. cURL / Python / Node.js 예제를 담은 상세 문서를 제공합니다.
APIMODELS는 같은 Grok 4.6을(를) 집약 플랫폼을 통해 제공합니다. API 인터페이스가 통합되어 있어 공급자마다 계정을 만들 필요가 없고, 키 하나로 모든 모델에 닿습니다.
$1.765 per 1M input tokens and $5.294 per 1M output, with cached input at $0.441. xAI official is $2.00 / $6.00, so both sides are 12% cheaper here. Pay-as-you-go, no minimum top-up, no subscription, and failed requests are never billed. Its sibling Grok 4.5 is $1.5 / $5 — cheaper, one capability tier down.
POST https://api.apimodels.app/v1/chat/completions with Authorization: Bearer YOUR_API_KEY and body {"model":"grok-4.6","messages":[{"role":"user","content":"..."}]}. It is plain OpenAI shape, so pointing openai-python or openai-node at base_url=https://api.apimodels.app/v1 works as-is, and so do Cursor, Cline and Codex. Add stream:true for streaming. /v1/responses is supported too.
The 4.6 release is aimed at long-running agents: tasks that span many steps — working across a codebase, turning an idea into a finished artifact — hold together better, and the model does more self-checking along the way. It scores 61 on the Artificial Analysis Intelligence Index, level with GPT-5.6 Sol, and beats 4.5 (56) on every published eval. Take Grok 4.5 ($1.5 / $5) when budget matters and tasks are short; take 4.6 for sustained work. Same endpoint — switching is a one-field change.
Yes, and we stress-tested it rather than smoke-testing it: non-streaming, multi-turn with tool results fed back, streaming, and parallel tool calls in one turn all passed 5/5. Then 12 concurrent calls against a nested schema with enums and arrays came back 12/12 correct — arguments parsed cleanly, no cross-request mixing. Grok upstreams have historically had intermittent tool-call failures on mixed pools; this channel did not reproduce that.
low, medium, high and xhigh, with high as the default. On one fixed prompt we measured 464 / 1,720 / 2,663 / 3,341 reasoning tokens respectively — strictly increasing, and unlike GPT-5.6, xhigh genuinely is the deepest tier for this model. The default level takes about 53 seconds, so set your client timeout above 60 seconds rather than leaving it at 30.
Because billed output tokens include reasoning tokens. A measured example: the model replied with a single number — one visible token — after spending 1,052 tokens reasoning, and the call bills 1,053. This is not padding on our side; the upstream charges us on exactly that basis and we pass it through. Lower reasoning_effort when you do not need the depth and the cost drops immediately.
Image input works — in testing it read text out of a picture accurately. Live web search runs on the **/v1/responses** endpoint with tools:[{"type":"web_search"}]; there it genuinely executes, returning web_search_call entries and real citation URLs. Passing search_parameters to /v1/chat/completions currently does nothing — the model emits the tool call as plain text instead — so do not rely on it on that path.
500K tokens. It is directly reachable from mainland China with no xAI account and no VPN, and the same API key also reaches every other model on the platform. Long conversations hit prompt caching automatically, billing the cached portion at $0.441 instead of $1.765; we also keep per-conversation routing sticky so repeat turns keep landing on a cache-warm server.
APIMODELS에서는 Grok 4.6이(가) 60개가 넘는 모델과 같은 API 키, 같은 잔액 위에 나란히 놓입니다. 그래서 선택은 궁합의 문제이지 종속의 문제가 아닙니다. Frontier、Long-Running Agents、Tool Calling、500K Context、Vision을(를) 지원하며 다른 대규모 언어 모델 모델과 가격·성능을 나란히 놓고 따져볼 수 있습니다. 갈아타기는 모델 이름 문자열 하나만 바꾸면 되고 새 계정도 추가 작업도 필요 없습니다. 대규모 언어 모델 선택지와 실시간 가격은 apimodels.app/models에서 볼 수 있습니다.
Grok 4.6은(는) 다음을 지원합니다: Frontier、Long-Running Agents、Tool Calling、500K Context、Vision. 전체 파라미터와 호출 예제는 APIMODELS 문서를 참고하세요.
네. APIMODELS는 Grok 4.6을(를) 하나의 통합 API와 키 한 개로 제공합니다. 공급자별 계정도 필요 없고, 각 공급자의 지역별 네트워크 경로를 직접 챙길 필요도 없습니다.
Stripe(Visa, Mastercard 등 해외 카드)와 Alipay를 지원합니다. 결제 후 잔액은 즉시 반영됩니다.