Grok 4.7 no Amazon Bedrock ao preço da xAI de 2/6 USD, e o escalão Flex reduz-o para metade
- O que aconteceu
- O Amazon Bedrock acrescentou o Grok 4.7 a 28 de setembro de 2026, ao preço de 2 USD/6 USD por 1 milhão de tokens na inferência entre regiões Global e a 2,20 USD/6,60 USD na US Geo.
- Porque é importante
- O escalão Flex do Bedrock fatura metade da tarifa Standard, um desconto que a API da própria xAI, sem suporte para processamento em lote, não oferece.
- O que fazer
- As equipas que usam AWS e têm trabalho não urgente com o Grok devem calcular o custo do Flex através das APIs Responses, Chat Completions ou Invoke, e testar primeiro a faturação de prompts longos.
Our verdict: Grok 4.7 on Bedrock is a real option for AWS shops, and the Flex tier is the reason to look. Global cross-Region pricing matches xAI's own $2/$6 per 1M tokens, and Flex cuts it to $1.00/$3.00 for work that is not time-sensitive. Our Grok 4.7 entry notes that xAI's own API does not support batch, so Bedrock is where the discount lives. Grok 4.7 stays conditional in our directory: that rating rests on benchmarks that are all xAI's own, and a new cloud does not change them.
What happened
AWS published the Bedrock model card for xAI's Grok 4.7 with a launch date of September 28, 2026 and an end of life no sooner than September 28, 2027.
- Context: 524,288 tokens, shared between input and output. Text and image in, text out.
- Standard pricing per 1M tokens: Global cross-Region $2.00 input, $6.00 output, $0.50 cache read. US Geo cross-Region $2.20, $6.60 and $0.55, a 10% premium to keep requests inside the US geography.
- Service tiers: Priority at 1.75x Standard, Flex at 0.5x. Reserved is not offered.
- Access: cross-Region inference only, via
us.xai.grok-4.7orglobal.xai.grok-4.7. In-Region inference is not supported. - APIs: Responses, Chat Completions, Converse and Invoke, on the OpenAI-compatible
bedrock-runtimeendpoint. - Reasoning: always on, with effort
low,medium,high(the default) orxhigh. The Chat Completions API does not return reasoning tokens. - Not supported: server-side tool use, intelligent prompt routing, token counting and explicit prompt caching. Implicit prompt caching is supported.
Why it matters
| Tier (Global cross-Region) | Input per 1M | Output per 1M |
|---|---|---|
| Standard | $2.00 | $6.00 |
| Flex (0.5x) | $1.00 | $3.00 |
| Priority (1.75x) | $3.50 | $10.50 |
The Standard rate on Global is the same headline price xAI charges directly, so moving to Bedrock costs nothing at list price and buys AWS billing, IAM and guardrails. The real gain is Flex: half price for background jobs, evals and bulk processing that xAI's own API, with no batch support, cannot discount.
Two limits narrow the fit. Flex and Priority run only on the Responses, Chat Completions and Invoke APIs; the card says Converse supports Standard only and returns a validation error for priority or flex. And with no in-Region endpoint, teams with single-Region residency rules cannot use Grok 4.7 on Bedrock at all; US Geo is the tightest option.
What changes for you
Check long-prompt billing first. Our entry records that on xAI's own API, prompts of 200K tokens or more bill at double the headline rate ($4/$12). The Bedrock card lists no separate rate for long prompts. We have not confirmed how Bedrock bills a 300K-token request, so test one before you budget long-context work on it.
Check your tool calls. Bedrock supports client-side tool calling only. If your agent relies on server-side tools, Bedrock is not a drop-in replacement for xAI's API: those calls have to run in your own code.
Pick the API before the tier. If your stack uses Converse, you are on Standard pricing. To get Flex, call Responses, Chat Completions or Invoke with "service_tier": "flex".
FAQ
Is Grok 4.7 cheaper on Bedrock than on xAI's API?
At Standard on Global cross-Region, no: both are $2/$6 per 1M tokens. It is cheaper only on the Flex tier, at $1/$3.
Can I keep Grok 4.7 traffic in one AWS Region?
No. The card lists in-Region inference as not supported. US Geo (us.xai.grok-4.7) keeps requests inside the US geography at a 10% premium.
Does Bedrock change our Grok 4.7 verdict?
No. Grok 4.7 stays conditional because every published score is xAI's own. Bedrock changes where you can run it and what you pay, not how good it is.
O que fazer
- 1 Use global.xai.grok-4.7 para a tarifa de 2 USD/6 USD, ou us.xai.grok-4.7 se os pedidos tiverem de ficar na geografia dos EUA, a 2,20 USD/6,60 USD.
- 2 Encaminhe o trabalho não urgente com o Grok 4.7 para o escalão Flex (service_tier flex) a metade da tarifa Standard, através da API Responses, Chat Completions ou Invoke, já que a Converse só suporta Standard.
- 3 Antes de mudar um agente para o Bedrock, verifique se depende de uso de ferramentas do lado do servidor, que a ficha do modelo no Bedrock indica como não suportado.
- 4 Envie um prompt com mais de 200 mil tokens e verifique a fatura antes de orçamentar trabalho de contexto longo no Bedrock.
Ferramentas e modelos afetados
Nunca mais precisas de te pôr a par
O resumo semanal: apenas mudanças de veredicto e ações urgentes. Sem enchimento.