## The New Frontier: OpenAI's Garlic, Apple's Clara, Real-Time Voice & Avatars Reshape AI

A recent wave of breakthroughs has swept across the artificial intelligence landscape, showcasing remarkable advancements from leading tech giants. OpenAI declared an internal "code red" and revealed a new model, Garlic, while Apple introduced Clara, a novel approach to document compression. Microsoft unveiled a real-time text-to-speech (TTS) model, Vibe Voice, effectively eliminating awkward delays in AI conversations. Alibaba presented Live Avatar, an avatar system capable of indefinite video streaming, and Tencent released Huan Video 1.5, a video generator accessible to a broader audience. These developments, occurring almost simultaneously, underscore a period of intense innovation and competition in the AI domain.

### OpenAI's "Code Red" and the Emergence of Garlic

The AI landscape recently saw an internal ["code red" declared within OpenAI](https://ingestgpt.com/blog/openai-code-red-vs-google-gemini-3), triggered by Google's Gemini 3 reaching the top of LLM Marina charts. This declaration by Sam Altman indicated significant pressure and that competition had closed the gap, necessitating an emergency response.

In this context, OpenAI has been secretly developing Garlic, a new model. Mark Chen, Chief Research Officer, confirmed Garlic's strong performance in internal tests, where it reportedly beats Gemini 3 and Anthropic's Opus 4.5 in reasoning and coding. These two models had previously set the gold standard in these categories.

Garlic's innovation stems from a rebuilt pre-training system. Instead of forcing the model to absorb all fine-grain details initially, the process now focuses on broader connections first, then narrows into specifics. This shift is described as major internally and allows packing more knowledge into smaller models, which are cheaper to train and faster to run. The rise of capable small models from Mistral, Deepseek, and Chinese labs influenced OpenAI's response.

Garlic is separate from Shallot Pete, another internal project aimed at fixing pre-training bugs, indicating OpenAI is running multiple model lines to leapfrog competition. While no official launch date, Mark Chen stated "as soon as possible," with early next year being a safe bet. The work on Garlic has already enabled progress on the next major model.

In contrast, Anthropic, according to Dario Amodei, does not compete for the same audience as OpenAI and Google, focusing on enterprise customers. Their Claude code system achieved a $1 billion revenue run rate within six months of launch.

### Apple's Clara: Rethinking Document Compression for AI

Apple quietly released Clara, a system designed to improve how AI handles long documents. Traditional chatbot systems often struggle with extensive texts, using methods that are slow, expensive, and less effective as documents grow. Clara addresses this by compressing documents into tiny memory tokens, which are super-dense summaries retaining important meaning. These tokens are then used for both retrieval and generation, allowing an AI to pull small bundles of compressed tokens instead of thousands of words.

Clara's innovative aspect includes training its retriever and generator together, allowing them to learn from each other, unlike most Retrieval Augmented Generation (RAG) systems that treat them separately.

The training involved approximately 2 million Wikipedia passages from 2021. A local Quen 32B model generated simple Q&A, multihop Q&A, and paraphrases. A verification loop of up to 10 rounds ensured data consistency. Two losses were used for training: cross entropy, to teach question answering with compressed memory, and MSE loss, to align memory and full document tokens. These details contribute to Clara's performance at high compression levels.

Clara's performance metrics are significant. At four times compression, its Compressor achieved an average F1 score of 39.86 on Natural Questions and Hot Pot QA, which is 5.37 points better than LLM Lingua 2 and over a point ahead of Pisco. In an Oracle setup, the model reached 66.76 F1, widely beating other methods. Compressed representations sometimes even outperformed full text retrieval pipelines like BGE plus Mistral 7B, delivering better results from shorter inputs. As a reranker, Clara achieved 96.21 recall at 5 on Hotpot QA, surpassing supervised retrievers.

Apple released three versions of Clara (Base, Instruct, and E2E) and the entire training pipeline, suggesting a potential larger push into the LLM space.

### Microsoft's Vibe Voice: The End of AI Speaking Delays

Microsoft introduced Vibe Voice Realtime 0.5B, a real-time voice model that resolves the awkward pauses in AI assistants. This model can begin speaking in about 300 milliseconds, essentially instantly.

It is designed for agents and LLMs that speak while thinking. As the language model generates text, Vibe Voice simultaneously converts tokens into speech, allowing the LLM to continue producing the rest.

The model uses an acoustic tokenizer running at 7.5 hertz, based on a sigma VAE system with seven transformer layers and a 3,200 times downsampling factor from 24 kHz audio. A small four-layer diffusion head sits on top, conditioned on hidden states from a Quen 2.5 to 0.5B model. The total stack is around 1 billion parameters.

Performance on Libre Speech Test Clean shows a 2% word error rate and a speaker similarity score of 0.695. Vibe Voice is optimized for long-form speech, maintaining stability across conversations, and can generate about 10 minutes of audio within its 8K context window. It keeps output clean for assistant-style use cases by not attempting music or background noise. Microsoft recommends running it as a small microservice alongside an LLM, with perfect synchronization between text and audio streams.

### Alibaba's Live Avatar: Infinite, Real-time Streaming Video

Alibaba, with major Chinese universities, released Live Avatar, an avatar system aiming to make animated avatars a real product. This system uses a 14 billion parameter diffusion model to generate video at over 20 frames per second in real time.

Users can speak into a mic, and the avatar responds instantly with smooth facial motion, gestures, and expressions. It supports streaming for over 10,000 seconds without drifting or losing identity.

This was achieved using several techniques. Distribution matching distillation shrunk a heavy multi-step video diffusion system into a model working with only four sampling steps. Timestep forcing pipeline parallelism spread the denoising process across GPUs for near-linear speedups, yielding an 84 times improvement.

The key innovation is solving the long video decay problem, where traditional auto-regressive video generators degrade over time (color shifts, face warping, identity drift). Live Avatar fixes this with three ideas:

**Rolling row PE**: Maintains stable positional information as video length increases.
**Adaptive attention sync**: Places the original reference frame with a generated one to prevent drift from the real distribution.
**History corrupt**: Adds controlled noise to the cache during training, teaching the model to recover from small mistakes.
Together, these elements allow Live Avatar to produce effectively infinite length streaming video without quality loss, a capability not seen before. Demos of two AI agents conversing with real-time avatars resemble sci-fi.

### Tencent's Huan Video 1.5: Accessible High-Quality Video Generation

Tencent introduced Huan Video 1.5, described as one of the most accessible, high-quality video generators this year. With only 8.3 billion parameters, a small size for video systems, it delivers top-tier video quality. This includes smooth motion, strong prompt following, clean text rendering, and stable camera movement. A key feature is its operational speed.

#### Conclusion

The recent cascade of advancements from OpenAI, Apple, Microsoft, Alibaba, and Tencent highlights a vibrant and intensely competitive era in artificial intelligence. From OpenAI's urgent internal "code red" driving the development of the high-performing Garlic model to Apple's innovative Clara for efficient document compression, Microsoft's Vibe Voice offering instantaneous AI speech, Alibaba's Live Avatar enabling infinite real-time video streaming, and Tencent's accessible Huan Video 1.5, these breakthroughs collectively push the boundaries of what AI can achieve. Each development, in its unique domain, signifies a substantial leap forward, promising to reshape how we interact with technology and process information in the coming years.

Explore the latest AI advancements: OpenAI's "code red" and the new Garlic model, Apple's Clara for document compression, Microsoft's instant Vibe Voice, and Alibaba's infinite Live Avatars.

AI's Rapid Evolution: OpenAI's Garlic, Apple's Clara, Real-Time Voice & Avatars

## OpenAI's "Code Red": The Battle for AI Supremacy Against Google's Gemini 3

The landscape of **artificial intelligence** is experiencing a seismic shift, with Google recently planting its flag firmly on the AI hill. Following the impressive capabilities demonstrated by **Gemini 3** and its proprietary **TPU architecture**, the perception of Google's lead in AI has grown significantly. This surge has prompted a decisive response from OpenAI, with CEO **Sam Altman** declaring a "Code Red" and confirming a new secret model, "Garlic," is in development to directly challenge Google's recent gains.

### Google's Formidable AI Position

Google's strengths in the AI race are comprehensive. The company possesses a frontier model, robust AI infrastructure, custom silicon in the form of TPUs, substantial revenue, a wealth of talented researchers, and an immense amount of data. This combination positions Google as a formidable competitor, especially after Gemini 3 showcased remarkable scaling capabilities, seemingly surpassing what many in the industry believed was a "wall" in pre-training.

The success of Gemini 3 is largely attributed to Google's TPUs, their custom silicon designed specifically for running artificial intelligence workloads. These TPUs have been in development for years, and their benefits are now evidently coming to fruition, enabling Google to deploy a massive model that performs incredibly well. This achievement led many to count OpenAI out, but the competition is far from over.

### The Scaling Wall Debate: A Crossroads for AI Research

Before Google's Gemini 3 demonstration, a consensus among leading AI researchers suggested that the "age of scaling" might be over. According to semi-analysis, OpenAI's researchers had not completed a successful full-scale pre-training run for a new frontier model that was broadly deployed since GPT-4o in May 2024, highlighting significant technical hurdles.

Ilya Sutskever, OpenAI co-founder and current founder of Safe Superintelligence, articulated this sentiment on the Dwarkesh Patel podcast. He stated that the industry is transitioning back to the "age of research," implying a need for new ideas and algorithms rather than merely scaling up existing methods. Sutskever noted that from 2012 to 2020 was the age of research, followed by the age of scaling from 2020 to 2025. However, he questioned whether a 100x increase in scale would still yield significantly different results, suggesting it might not.

Andre Karpathy, another prominent AI researcher, echoed similar views on the same podcast, while Yann LeCun has consistently maintained for years that large language models in their current form are reaching their limits and require new fundamental ideas. These perspectives underscored a belief that LLMs were at a dead end and that simply scaling them further was no longer effective.

However, Gemini 3 challenged this notion, demonstrating that scaling is "far from dead." This has led to speculation that Google's unique architecture or methods, particularly their TPUs, allowed them to overcome the perceived pre-training scaling wall.

### OpenAI's "Code Red" and Strategic Reorientation

In response to Google's advancements and the industry's shift, Sam Altman declared a "Code Red" within OpenAI. This directive signifies an intense focus on improving the quality of ChatGPT and competing fiercely in the grand model race, even at the sacrifice of progressing other products. Internal memos, such as the Kobayashi letter, indicate that OpenAI is pausing projects like ads and shopping to concentrate on enhancing ChatGPT's performance.

Altman explicitly stated that OpenAI has "more work to do on the day-to-day experience of its chatbot." Key areas of improvement include enhancing personalization features for users, increasing speed and reliability, and enabling the chatbot to answer a wider range of questions. This strategic shift emphasizes that for the vast majority of users and use cases, the core intelligence of current models is already sufficient. What truly matters is the "day-to-day experience" and the surrounding "scaffolding" that makes the AI accessible and user-friendly.

For 99% of users, the models are good enough to serve their needs. ChatGPT has maintained its position as "the verb" for AI interaction, similar to how Google became synonymous with search. The ChatGPT experience is still considered "killer," and despite some frontier users switching to whichever model is momentarily superior, the majority of the population is unlikely to switch unless major issues arise with ChatGPT.

### Introducing "Garlic": OpenAI's New Frontier Model

As part of its "Code Red" initiative, **OpenAI** is actively developing a new model code-named "Garlic." This model is specifically designed to counter Google's recent gains, particularly on the pre-training front. OpenAI's Chief Research Officer, **Mark Chen**, recently confirmed a supercharging of pre-training efforts over the last half-year, with a dedicated "superstar team" focusing on building this crucial muscle at OpenAI.

Chen explicitly stated that OpenAI believes there is "a lot of room left in pre-training" and that scaling is "not dead at all." He expressed confidence that these renewed efforts are leading to the training of "much stronger models," allowing OpenAI to go "head-to-head with Gemini 3 easily on pre-training."

"Garlic" has reportedly been performing well on OpenAI's internal evaluations when compared to Gemini 3 and Anthropic's Opus 4.5. It's suggested that "Garlic" incorporates bug fixes developed during the pre-training process of another rumored model, "Shallot Pete." Furthermore, Altman hinted at a potential "December surprise," which could mean the release of GPT 5.2 or even 5.5 by the end of the year, although "Garlic" is distinct from "Shallot Pete."

### The Competitive AI Landscape

The AI market is intensifying. Google has seen significant user growth, expanding from 450 million active users to 650 million in just a few months. Google is also incredibly well-positioned financially, with diverse revenue streams and substantial cash reserves, allowing them to fund and potentially absorb losses in their AI division while competing with OpenAI and Anthropic.

OpenAI, while nearing a billion active users, has reportedly seen its growth slow down. Despite this, the company remains a formidable force. The intense competition drives innovation, pushing both giants to refine their strategies and technologies.

For those navigating this complex and rapidly evolving AI landscape, accessing and comparing different models can be challenging and expensive. Chat LLM by Abacus AI offers a solution, providing access to multiple leading AI models such as **ChatGPT 5**, O3 Pro, Gemini 3 Pro, Claude 4.5, Grok 4, **Deepseek**, and others, all in one place for just $10 per month. Its "Route LLM" feature automatically selects the best model for a given prompt, and the "Humanize" feature can transform AI-generated text into more compelling content. Additionally, their "Deep Agent" offers a powerful general agent capable of building websites, apps, AI agents, chatbots, presentations, and research reports.

### Conclusion

The AI race between Google and OpenAI has entered a new, highly competitive phase. Google's Gemini 3 and **TPU architecture** have challenged previous assumptions about the limits of AI scaling, prompting OpenAI's swift "Code Red" response. With Sam Altman's focus on enhancing **ChatGPT's** user experience and the active development of powerful new models like "Garlic," OpenAI is signaling its intent to reclaim its lead. The debate over scaling versus research continues, but one thing is clear: the pursuit of advanced artificial intelligence is pushing the boundaries of innovation, promising exciting developments for users and the industry alike.

Sam Altman declares "Code Red" at OpenAI, focusing on a new model, "Garlic," and user experience to compete with Google's Gemini 3 and its TPU architecture. Explore the shifting landscape of AI scaling and research.

Category: AI Models

AI's Rapid Evolution: OpenAI's Garlic, Apple's Clara, Real-Time Voice & Avatars

Inside OpenAI’s Code Red: The Secret Plan to Outrun Google’s Gemini 3