Billion leaf Elon Musk has Now has been counted by OpeniBut it doesn’t stop from its AI company, Zai, Making its flagship Grook 3 model available via API.
It has been several months to the masking of Zai Grook 3Company’s response to models like Openi GPT-4O And Google’s Gymnony. Grook can analyze 3 images and answer questions, and Mask’s social network, X -to -X -to -X -to -X -to -X -has -the Zee not in March, not co -coining.
Zee is offering two flavors of his flagship model through his API: Grook 3 and Grook 3 with “reasoning” capabilities.
Grook 3 is priced in a $ 3 million token (~ 750,000 words) model and a token of $ 15 million produced by the model. Meanwhile, Grook 3 mini will cost $ 0.30 per million input token and 50 0.50 per million output token.
The sharp versions of Grook 3 and Grook 3 mini are available on a premium: GROK 5 for 5 million input token and $ 25 per million output token Grok 3. And Grok 0.60 per million input token and GROK 3 MINI for $ 4 million output token.
Grook is not cheaper than 3 competition. Zee is matching anthropic prices Claude 3.7 SwantWhich also offers the reasoning capabilities, but it is far more expensive than Google’s recent release. Gemini 2.5 ProWhich usually receives more than 3 scores in the popular AI benchmark. (Specifically, Zai has been charged Being misleading In his Grook 3 benchmark reports.)
As if Multiple users on x indicatedGrook 3, through Xai’s API, also has a small context window that is considered to be worth supporting the model. (The “context window” refers to how many tokens can be followed at the same time.) API 131,072 tokens, or about 97 97,500 words – is less than one million tokens, Xi claimed that Grook 3 was supported at the end of February.
When Musk announced Grook almost two years ago, he made the AI model as sharp, unprecedented and anti-“awake”-generally, ready to answer controversial questions that other AI systems would not. He provided some of these promises. Been told to be unconscious, for example, Grook and Grook 2 will be binding it with happiness, the possibility of giving birth to the colorful language you won’t listen to Chat GPT.
But the Grook model before the Grook 3 Hedge On political subjects and will not cross Some limits. In fact, A study It is learned that Grook has been leaning on topics like transgender rights, diversity programs and inequality.
Kasturi has blamed this behavior of Grook’s Training Statistics – Public Web Pages – and Promised “Shift Grook near politically neutral.” Lack of high -level errors Briefly censored President Donald Trump and Kasturi’s MentionIt is not yet clear whether Zee has achieved this goal at the model level, and what could be long -term consequences.