🔥Tencent Unveils DeepSeek Rival

Today we will discuss:

• Sponsored: PostHog—helping engineers and founders build better products with insights, lessons, and strategies

• Tencent’s Turbo S vs. DeepSeek R1—Tencent claims its new AI model responds quicker and costs less

• Diffusion meets LLMs—Inception’s new approach promises faster, more efficient AI models

• In other AI news today—ElevenLabs launches new speech-to-text model Scribe, Perplexity AI announces a $50M fund for early-stage startups, Baidu plans mid-March release of Ernie 4.5

• Don’t miss this week’s Workflow Wednesday—AI monetization made easy with ScribeHow, Pictory, and Gravity Write; now available at only $20/month! 

• Saved the best for last—9 must-try AI tools

Product for Engineers: Flex your product muscles

Product for Engineers is PostHog’s newsletter dedicated to helping engineers and founders improve their product skills. In it, they provide curated advice on building great products, lessons (and mistakes) they’ve learned building PostHog, and deep dives on the culture and strategies of top startups. Subscribe for free.

Tencent unveils Turbo S, says it’s faster than DeepSeek R1

Key Points 

• Tencent says its new AI model, Turbo S, can generate responses in under a second, outperforming DeepSeek R1.

• The company claims Turbo S matches DeepSeek-V3 in knowledge, math, and reasoning.

♨️News - Tencent has unveiled Hunyuan Turbo S, a new AI model it claims can respond faster than DeepSeek’s R1, a model that has gained significant attention in the AI space.

🧐Does it measure up? According to Tencent, Turbo S can generate responses in under a second, setting it apart from models like DeepSeek R1 and Hunyuan T1, which take longer to process before answering. 

The company also says that in areas like knowledge, math, and reasoning, Turbo S performs at the same level as DeepSeek-V3, the model behind DeepSeek’s AI chatbot, which recently surpassed OpenAI’s ChatGPT in app store downloads.

Tencent is also emphasizing cost efficiency, saying Turbo S is significantly cheaper to use than its previous versions. The move reflects a growing trend in China’s AI industry, where DeepSeek’s open-source approach and low pricing have forced major players to rethink their pricing strategies.

Read the full story here. 

New AI startup Inception introduces diffusion-based text model to challenge traditional LLMs

Key Points 

• Unlike LLMs, which generate words sequentially, Inception’s model refines entire text blocks in parallel for greater efficiency.

• The startup claims its AI models run 10x faster than competitors while reducing GPU costs for enterprise users.

☕News - A new AI startup, Inception, is making bold claims about a faster, more efficient approach to language models. Founded by Stanford professor Stefano Ermon, the Palo Alto-based company has introduced a “diffusion-based large language model” (DLM), applying diffusion technology—commonly used for images and video—to text generation.

👨‍💻For context - Most generative AI models today fall into two categories: large language models (LLMs), which generate text, and diffusion models, which create images, video, and audio. While LLMs like ChatGPT process words sequentially, diffusion models work differently, refining an entire output in parallel.

Inception’s DLM is designed to offer the same core capabilities as LLMs—code generation, answering questions, and more—but with significantly faster performance and lower computational costs.

✨The unique approach - Ermon’s research focused on overcoming a key limitation of LLMs: their step-by-step generation process. “You cannot generate the second word until you’ve generated the first one,” he explained. Diffusion models, however, start with a rough estimate and refine everything simultaneously.

After years of research, Ermon and a student developed a way to apply this approach to text. Their breakthrough, published last year, demonstrated that generating and modifying large blocks of text in parallel was possible—potentially reshaping how language models are built.

🤓What it's delivering - According to Ermon, this method allows models to use GPUs far more efficiently, making AI systems both faster and cheaper. Inception claims its DLMs run up to 10 times faster than traditional LLMs while cutting costs by the same factor.

The company is already offering an API, on-premises and edge deployment, and model fine-tuning options. A spokesperson compared Inception’s smallest coding model to OpenAI’s GPT-4o mini, saying it delivers similar results but at more than 10 times the speed.

Read the full story here. 

We just launched a new edition of Workflow Wednesday for AI-minded professionals like you. Actionable AI-workflows delivered directly to your inbox.  This week's Topic? AI-Monetization

Join Here