xAI Launches Grok 4.6

In partnership with

Grok 4.6 shipped five weeks after Grok 4.5

SpaceXAI, the company formerly known as xAI, released Grok 4.6 on 12 August. It scored 61 on the Artificial Analysis Intelligence Index, level with GPT-5.6 Sol and just behind Claude Opus 5, while holding price flat at $2 per million input tokens and $6 per million output.

The 500,000 token context window did not change, which is now the smallest among frontier models. Its real edge is turn efficiency: it finishes agent tasks in roughly half the steps. It went live the same day in Cursor, Grok Build, and the API. No system card was published. ^{_{Source: x.ai}}

Brands Publish 9.5 Social Posts a Day. This AI Writes and Publishes Yours in Your Own Voice

Your voice. Every platform. No writing required.

You ghost your own socials by Wednesday. SureThing learns your voice and ships native posts to LinkedIn, X, Instagram, and TikTok, without you writing a thing.

Post in Your Voice

How to Create Funny Cartoon Using Pippit agent

Claude just moved a 167 year old math problem further than humans managed in 37 years

Anthropic announced on 10 August that an unreleased research version of Claude raised the proven lower bound for Riemann zeta zeros sitting on the critical line from 41.6% to 67.2%. It did not prove the Riemann Hypothesis, and Anthropic says this approach never will. What it did do is jump a number that had moved 0.8 points in the previous 37 years. The run used roughly 60 subagents inside Claude Code, 2,400 shell commands, and 650 failed ideas before one worked. Two in house mathematicians checked it, Brian Conrey and Dan Goldston reviewed it, and the Lean 4 formalization is public on GitHub. That last part is why this reads differently from the usual AI math announcement. ^{_{Source: Anthropic}}

ChatGPT Secret Codes For Marketers

invideo agent two creates insane video clips without writing prompts

5 AI Tools To Start 5 AI Side Hustles

AI & Tech News

AI writes more code, and CodeRabbit raises $143 million to check it

AI code review company CodeRabbit raised $143 million in Series C financing at a $1.5 billion valuation, with Atomico and Smash Capital co leading the round. The company says it now performs more than two million code reviews a week for over 17,000 customers, including Nvidia, BMW, JFrog and Indeed.

Alongside the funding, CodeRabbit introduced what it calls Agentic Change Management, designed to validate, explain, prioritise and monitor software changes produced by both people and coding agents.

The underlying industry shift is straightforward: as AI makes producing code faster and cheaper, reviewing that code for quality, security and business value becomes a bigger part of the engineering workload. ^{_{Original source: coderabbit.ai}}

Meta released Muse Glimmer

A 30 billion parameter open weight model under Apache 2.0 that runs on a single consumer GPU. Same day, Zuckerberg published a 6,500 word essay called The Future Is for Everyone arguing that concentrating superintelligence in a few companies is the real risk. He also pledged to open the weights of Muse Spark 1.2, announced a $1 billion fund for communities near Meta data centers, and asked Washington to stop making open models harder to build in the US. The timing was not an accident. Meta spent most of 2025 walking away from open releases, and this is the reversal. ^{_{Source: Meta}}

Everything Claude writes now carries an invisible watermark

Anthropic confirmed that Claude models launched on or after 2 August embed a hidden statistical watermark in generated text, plus signed C2PA provenance data on image files. It applies worldwide, not just in the EU, and there's no opt out. The trigger is Article 50 of the EU AI Act, which became enforceable on 2 August. The mark survives copy and paste and may survive light editing. Worth knowing: Anthropic says a detected mark means Claude processed the text, not that Claude wrote it. Proofread your own paragraph through Claude and it can pick up the signal. s^{_{ource: Techcrunch}}

A four month old startup raised $1.1 billion on the idea that you should own your AI

River AI, founded by xAI co founder Igor Babuschkin, closed $1.1 billion on 11 August across seed and Series A. General Catalyst and AMP PBC led it. Nvidia, AMD Ventures, Y Combinator, and Temasek all joined. The pitch is that companies should train and own models on their own data rather than renting general purpose ones, and River sells an API for fine tuning and reinforcement learning on open weight models. It claims a full RL training run takes 15 to 20 minutes with no infrastructure team. The company has about 20 employees in Palo Alto and has not shipped its open source stack yet. _{^{Source: Techcrunch}}

Nvidia put out a free model that runs on one GPU, and the reason is obvious

Nemotron 3.5 Lightning, It's a 30 billion parameter mixture of experts model with only 3 billion active at a time, built for the boring high volume steps in agent workflows: tool calls, checking results, delegating to subagents. Nvidia reports up to 4x faster output than similar sized models and a 1M token context window, running on a single H100 or DGX Spark. Weights, training data, and recipes are open under a permissive license. Jensen Huang said the quiet part out loud in July: free AI is good for chips, because free models still need somewhere to run. _{^{Source: cnbc}}

ChatGPT ads went international

ChatGPT Ads has launched in the United Kingdom, Mexico, Brazil, Japan, and South Korea, with more markets coming this year. Ads run only for logged in adult users on the Free and Go tiers. Paid tiers stay clean. OpenAI says ads sit below responses, carry sponsored labels, and don't influence what the model tells you. Two years ago Sam Altman called ads in AI a last resort. The economics of serving hundreds of millions of free users appear to have settled that debate. ^{_{Source: Openai}}

Gemini 3.7 Flash goes after the everyday AI workload

Google launched Gemini 3.7 Flash, positioning the model for software development, automated business tasks and agents that carry out multistep workflows using software tools. Google said the new model improves software engineering, debugging and production code work while retaining the lower cost profile expected from its Flash family.

Introductory API pricing is $0.75 per million input tokens and $3.75 per million output tokens through the end of 2026, half the original Gemini 3.6 Flash price.

Google also began putting 3.7 Flash into Gemini Spark immediately, making this a commercial deployment rather than a research preview. _{^{Source: Google}}

OpenAI's best model now runs at 750 tokens per second

OpenAI previewed Ultrafast, an API tier running the full GPT-5.6 Sol on Cerebras wafer scale hardware at up to 750 output tokens per second. That's up to 14 times the standard serving speed, with no smaller or distilled model involved. Cerebras keeps all 44GB of weights on chip, which removes the memory shuffling that slows GPUs down. Jane Street, Podium, Basis, and Rogo are testing it across voice, coding, and financial research. It's limited preview only, with no published price and no GA date. OpenAI engineers are reportedly using it to read logs during incident response. ^{_{Source: Openai}}

Top AI tools to check out…

I’ve been building a best AI tool directory where you’ll find the best of the best tool and their USECASES.

Thanks for your time!

Shailesh & OpenAILearning team