Julian

[Artificial Analysis @ArtificialAnlys]

Grok 3 mini Reasoning’s recently launched API has highly compelling Intelligence vs. Price positioning

@xai has recently launched APIs for Grok 3 and Grok 3 mini, after initially only making the models available via the Grok chat interface at launch. We have now completed benchmarking for both Grok 3 and Grok 3 mini.

Grok 3 mini (high reasoning) stands out in the top left of our Intelligence vs Price chart, achieving one of the highest Artificial Analysis Intelligence Index scores ever, with pricing well below even DeepSeek R1.

Grok 3 family overview: We’re initiating coverage of 6 versions of the Grok 3 family: Grok 3, Grok 3 Fast, Grok 3 mini (low reasoning), Grok 3 mini Fast (low reasoning), Grok 3 mini (high reasoning) and Grok 3 mini Fast (high reasoning). Today’s post focuses on our intelligence results; performance breakdown of all 6 versions will be available soon.

Intelligence: As we highlighted at launch, Grok 3 marks xAI’s arrival at the AI frontier. Both Grok 3 mini Reasoning (reasoning effort high) and Grok 3 are amongst the top 5 models in their reasoning and non-reasoning categories respectively. Grok 3 mini Reasoning (high) ranks higher in our Artificial Analysis Intelligence Index than DeepSeek R1 and Claude 3.7 Sonnet (64k reasoning budget).

Pricing: Grok 3 mini Reasoning is almost an order of magnitude cheaper than other models of similar intelligence (o4-mini, Gemini 2.5 Pro) at $0.3/$0.5 per 1M Input/Output tokens for the ‘base’ speed version. xAI are charging $0.6/$4 for the faster Grok 3 mini endpoint. OpenAI’s o4-mini is $1.1/$4.4 and Gemini 2.5 Pro $1.25/$10. Grok 3 is more expensive at $3/$15 per 1M Input/Output tokens.

End to End Latency: Grok 3 offers significantly faster response times compared to Grok 3 mini Reasoning due to ‘thinking’ time. We measure Grok 3 as taking 9.5s to output 500 tokens, and Grok 3 mini Reasoning as taking 27.4s to complete reasoning and output 500 tokens (standard speed endpoints, comparison to Fast endpoints coming soon).

[v0 @v0]

With AI integrations, you can access the @xai free tier in v0, exclusive to Vercel. Get $10 to use across all Grok models.

[Julian @brightsparc]

It's not too late to spend big on Kubernetes 😂

[Simon Wardley @swardley]

X : Predictions for 2019?
Me : Not my thing.
X : Go on.
Me : What are you after?
X : Acquisitions.
Me : Hmmm, a bucketload of tech acquisitions around Kubernetes. Expect big juicy deals.
X : It's the future?
Me : No. Mostly cash rich past giants with no future being clueless.

[Francisco Massa @fvsmassa]

Today we are releasing Mask R-CNN Benchmark: a fast and modular implementation for Faster R-CNN and Mask R-CNN written entirely in @PyTorch 1.0. It brings up to 30% speedup compared to mmdetection during training. Check out the webcam demo! https://t.co/LQ2YKYRO3N

[Julian @brightsparc]

👋🏻 @deeplens_selfie

[Eric Topol @EricTopol]

Impressive prospective demonstration of how deep learning #AI plus pathologists is better than either component alone (faster, easier, more accurate breast #cancer metastasis diagnosis) @AJSPjournal https://t.co/R1vIzVJcGI #OA by @GoogleAI @davesteiner @lhpeng and colleagues

[The Economist @TheEconomist]

In Britain, half a million people regularly go up to a week without seeing anybody #OpenFuture https://t.co/WnC9CMdjx1

[Jeremy Howard @jeremyphoward]

fastai's docs for the upcoming v1 are fully generated from jupyter notebooks using a new documentation framework we'll be releasing, which even automatically hyperlinks all symbols

[Susan Etlinger @setlinger]

Nice interview with @RichardSocher on current capabilities and future potential of #AI, plus a reminder to focus on the immediate #ethics issues with #bias in data rather than all the SciFi stuff. https://t.co/zvqnOqnfAv

[Jeremy Howard @jeremyphoward]

Congrats to @nirantk for his state of the art in Hindi language modeling and text classification using our ULMFit approach (w @seb_ruder).
https://t.co/51O1ylCZaT

[Ai2 @allen_ai]

#AI2's model #ELMo "swiftly toppled previous bests on a variety of challenging tasks ... In a field where progress tends to be incremental, adding ELMo improved results by as much as 25%." Challenges & successes in #MachineComprehension: https://t.co/EHhnZHU8Yq via @WIRED #AI

[Sebastian Ruder @seb_ruder]

Succinct and sober @WIRED article about recent work on pre-trained language models from @OpenAI, @allen_ai, @fastdotai, and me. Sidesteps hype by outlining not only successes but also risks. https://t.co/Lj5Fsb13gJ

[WIRED @WIRED]

Russia is using Ukraine as a cyberwar training ground. The United States could be next. https://t.co/isqlzYczGZ

[Jeremy Howard @jeremyphoward]

Fantastic to see Google really starting to care about dev UX, and using the great work from @fchollet as their foundation

[Andrew Cantos @andrewcantos]

Salesforce open-sources TransmogrifAI, the machine learning library that powers Einstein https://t.co/ULPLgzZFIy

[WIRED @WIRED]

NASA is now preparing for a human mission to Mars, but @jason_pontin says if our descendants ever shrug off their terrestrial bonds, it won’t be Homo sapiens who leave, but another, more intelligently designed species. https://t.co/pZRPHUUCVp

[WIRED @WIRED]

We talk a lot about all kinds of exciting transportation technology like drones, self-driving and flying cars, and Hyperloop. But the most exciting one? Well it's the good old fashioned bicycle. https://t.co/kHEWVt8JZN

[Jeremy Howard @jeremyphoward]

I haven't tried this yet myself, but it looks quite hopeful

[Andrew Cantos @andrewcantos]

AI: What's Working, What's Not https://t.co/wPyjzPk3qB