What's next? With the Grok 3 API now live, xAI is stepping directly into the ring with OpenAI, Anthropic, and Google. Pricing signals ambition, but performance, transparency, and adoption will ultimately decide who wins.
GENERATIVE AI
-
Grok’s unfiltered promise questioned after left-leaning tests
By
–
A quick political detour: Remember Grok’s edgy, unfiltered promise? Musk said it wouldn’t be “woke.” But earlier Grok models leaned left in tests. It’s unclear if Grok 3 corrected that or simply rebranded it.
-

Grok 3 API context limit maxes at 131K tokens, not 1M
By
–

Some early catches: The API maxes out at 131K tokens,
not the 1M tokens xAI previously claimed. Users on X have already pointed it out. Also, Grok 3’s context behavior seems narrower than advertised. -
Grok 3 pricing and benchmark concerns versus Claude and Gemini
By
–
It’s not the cheapest model on the market. Grok 3 is on par with Claude 3.7 Sonnet, but pricier than Google’s Gemini 2.5 Pro. And benchmark results? Let’s just say xAI’s reporting has raised a few eyebrows.
-

Grok 3 Pricing Breakdown for Standard, Speed, Mini
By
–
Here’s the full pricing breakdown: Grok 3 Standard • $3M / input tokens
• $15M / output tokens Grok 3 Speed • $5M / input
• $25M / output Grok 3 Mini Standard • $0.30M / input
• $0.50M / output Grok 3 Mini Speed • $0.60M / input
• $4M / output -

Grok 3 versions: Full model and Mini, each in Standard and Speed.
By
–
There are 2 versions: • Grok 3 (full model)
• Grok 3 Mini (lighter, cheaper) Each one comes in Standard and Speed editions. -

xAI launches Grok 3 API: fast, expensive, real-world ready
By
–
BREAKING: Elon Musk’s xAI just launched the API for Grok 3. It’s fast, expensive, and ready for real-world use. Here’s everything you need to know
-
Seed-Thinking-v1.5: Advanced AI Model Capabilities Showcase
By
–
Now, can you imagine how strong Seed-Thinking-v1.5 will be?
Seed-Thinking-v1.5: https://
github.com/ByteDance-Seed
/Seed-Thinking-v1.5/blob/main/seed-thinking-v1.5.pdf
…
DAPO: https://
arxiv.org/abs/2503.14476
VAPO: -

Huawei Pangu Ultra 135B Matches DeepSeek-R1 Performance
By
–
Huawei is gearing up to release Pangu Ultra—a 135B dense Transformer trained on 13.2T tokens with 8,192 Ascend NPUs. Despite fewer parameters, it matches DeepSeek-R1 in performance.
No full tech report or model yet. Check it out: https://
arxiv.org/abs/2504.07866 -

Google begins limited rollout of Veo 2 on AI Studio
By
–
BREAKING : Seeing multiple reports that Veo 2 is rolling out on Google AI Studio. Seems like it is a very limited rollout for now. h/t @HarisHiew