By Mulubwa Chungu (Technical Lead at BongoHive Consult – Backend & DevOps, Gen AI Core Team)

This past week has been another whirlwind for artificial intelligence. In just a few days, we saw Meta unveil a standalone AI app and free Llama API preview, Nvidia push reasoning benchmarks higher, Google aims its models at dolphin communication, and Elon Musk teases Grok 3.5. 

Meanwhile, DeepSeek, OpenAI, Alibaba, and Perplexity all shipped notable updates. The AI race is accelerating on every front.

Meta’s LlamaCon 2025 Ushers In a Stand-Alone AI App and Free Llama API
Meta used its first-ever LlamaCon to launch a dedicated “Meta AI” mobile app to rival ChatGPT, open a free preview of the Llama API, and debut new safety tools such as Llama Guard 4, LlamaFirewall, and Prompt Guard. It also announced inference partnerships with Groq and Cerebras, promising speeds up to 18× faster than standard GPU stacks. Read More

Nvidia’s Nemotron Ultra Raises the Reasoning Bar
Nvidia followed its earlier Nemotron-4 family with the new 253 B-parameter “Nemotron Ultra,” showing stronger multi-step reasoning than DeepSeek R1 while staying half the size. It also topped LiveCodeBench for real-world coding tasks.

Google’s DolphinGemma Tries to Decode Dolphin Speech
Google DeepMind unveiled DolphinGemma, a 400 M-parameter on-device model trained on decades of dolphin vocalizations. The company plans an open-source release this summer to help biologists study inter-species communication.

Grok 3.5 Beta Promises First-Principles Answers
Elon Musk confirmed that Grok 3.5 enters beta for SuperGrok subscribers next week, focusing on answers “that don’t exist on the internet” by reasoning from first principles. Microsoft is already preparing to host Grok in Azure AI Foundry.

DeepSeek’s Open-Source R1 & V3 Keep the Pressure On
Chinese startup DeepSeek upgraded its V3 model and is hiring aggressively to productize its low-cost reasoning engine, while its fully open-source R1 continues to perform near proprietary peers at a fraction of the price.

OpenAI Releases GPT-Image-1 to Developers
OpenAI quietly pushed its gpt-image-1 multimodal generator into the public Images API, and design suites like Adobe Firefly and Figma are already integrating it.

Alibaba Announces Qwen 3 Hybrid Reasoning Models
Alibaba introduced Qwen 3;  a mixture of expert models up to 235 B parameters and six dense models claiming parity with Google and OpenAI on coding and math.

Perplexity Voice Assistant Lands on iOS and Motorola
Perplexity rolled out its multimodal voice assistant to iPhone and iPad, then struck a deal to pre-install the assistant on Motorola’s upcoming Razr, signaling a push into hardware search.

In a week, AI innovation surged across reasoning models, multimodal research, and everyday assistants. Meta doubled down on openness, Nvidia and DeepSeek traded blows on benchmarks, Google aimed AI at the natural world, and Grok 3.5 upped the stakes in first-principles reasoning. 

With OpenAI, Alibaba, and Perplexity expanding their ecosystems, intelligent capabilities are weaving deeper into both consumer and enterprise tools at breakneck speed. 

The takeaway? AI’s evolution isn’t slowing, it’s compounding. 

To find out more about what are doing in AI, visit :ai.bongohive.co.zm

To explore what these trends mean for your organization, reach out at :[email protected].