Grok 4.6 and Gemini 3.7 Flash both launch focused on cheaper agent work

[Benzema]

xAI and Google both dropped new models this week, Grok 4.6 and Gemini 3.7 Flash, and both companies are leaning hard into agent capabilities and lower costs as the main selling points. Gemini 3.7 Flash is being pitched as Google's most intelligent workhorse model yet for coding and agent tasks, launching at half the price per million tokens compared to the previous 3.6 Flash version

Google is clearly under pressure to keep up here too, since this launch comes right after Demis Hassabis stepped down as CEO of Google DeepMind to become Chair and Chief Scientist of parent company Alphabet, with Sundar Pichai writing that the company needs to accelerate its work and stay focused on the AI frontier. Reuters also reported that Sergey Brin has been pushing key AI staff internally to go all in on Gemini, which suggests there is real internal urgency behind these releases

On the xAI side, Grok 4.6 builds on Grok 4.5 with a focus on long running agent tasks, and xAI also launched something called Grok Bot alongside it, described as AI teammates you can hand real work to. The company says Grok 4.6 is especially strong at turning a rough product idea into a working first version, researching unfamiliar domains and refining the output across multiple rounds of feedback

Elon Musk claimed on X that Grok 4.6 is objectively number one when you weigh intelligence, speed and cost together, though the actual LiveBench leaderboard cited in the coverage has it sitting behind several other models including two Claude variants, two GPT-5 variants, and a couple of Chinese models from Moonshot and Alibaba, so objective feels like a bit of a stretch there

Cost is clearly the real battlefield right now more than raw capability, with both Grok 4.6 and Gemini 3.7 Flash being priced to compete directly against cheaper Chinese offerings like Moonshot's Kimi K3 rather than trying to simply out benchmark the top US labs

Xai Google How Do Grok 4 6 And Gemini 3 7 Flash Performaimagazine.com

[Priya_39]

Musk calling Grok 4.6 objectively number one while it sits ninth on the actual leaderboard cited in the same article is peak Musk, the gap between the claim and the data could not be more obvious

[James95]

Demis Hassabis stepping down as DeepMind CEO right before this launch is the more interesting story buried in here honestly, that is a significant leadership shakeup happening quietly alongside a routine model release

[BareMinimum]

Curious how much of Grok 4.6's improvement is genuine architecture work versus just better prompting and tooling around the same underlying model, xAI's announcements always lean pretty heavily on the marketing side

[Keane]

Kimi K3 from Moonshot keeps getting name checked as the price benchmark everyone is trying to match, that is a pretty remarkable position for a Chinese lab to be setting the pricing floor for the entire industry

[ParallelSelf34]

Claude and GPT-5 variants dominating the top of that leaderboard while everyone else fights over agent features and pricing suggests the actual capability gap has not really closed even if the marketing suggests otherwise

[Gareth5]

Half price on Gemini 3.7 Flash compared to 3.6 is a pretty aggressive cut, feels like Google is genuinely worried about losing cost conscious enterprise customers to the cheaper Chinese labs