Getting Rid of NVIDIA Dependency! OpenAI Joins Forces with Cerebras to Launch GPT-5.3-Codex-Spark: The First Fruit of a $10 Billion Computing Power

To gain more initiative in the computing power battlefield, OpenAI is accelerating its "de-NVIDIA" strategy. On February 12, 2026, OpenAI officially launched its first AI model based on Cerebras Systems chips - GPT-5.3-Codex-Spark.

The "Lightning" Evolution of Programming Experience

As a productivity tool specifically designed for software engineers, GPT-5.3-Codex-Spark brings a more flexible interaction mode:

Instant Interruption and Switching: Developers can interrupt long-running computational processes at any time, without waiting, and instruct the model to switch to handle other urgent coding tasks.

Full-Process Coverage: The model can assist engineers in quickly completing high-frequency operations such as editing specific code segments and running automated tests.

A $10 Billion Bet on Non-Standard Computing Power

This model release marks the first milestone after OpenAI and Cerebras entered into a deep partnership.

Expensive Agreement: The two parties have signed a multi-year agreement worth over 10 billion dollars, aiming to achieve ultra-large-scale computing power deployment of 750 megawatts.

Long-Term Strategy: The related production capacity will be rolled out in phases, with the full hardware deployment expected to be completed by 2028.

Industry Insight:

As AI competition enters deeper waters, a single chip supply channel has become a bottleneck restricting the development of major players. By supporting Cerebras and running core models on its hardware, OpenAI not only achieved breakthroughs at the technical level but also secured a "card" to free itself from NVIDIA's constraints at the strategic level.

Spring Festival Orders Expected to Exceed 5,000! Qingtian ZHU GMV Increased by 80% Quarter-over-quarter

During the Spring Festival, the robot rental platform Qingtian ZHU saw a surge in orders, with expected holiday orders exceeding 5,000. The overall GMV increased by about 80% quarter-over-quarter. Order growth remained strong before and after the holiday. About 30% of first-time renters indicated an increase in market penetration.

Meitu Kuaipai Joins Seedance 2.0 Large Model in Its First Batch

The video recording tool 'Kuaipai' under Meitu will integrate the Seedance 2.0 large model, expected to launch at the end of February. This move aims to deeply embed AI generation capabilities into users' workflows, improving the efficiency of short video creation. As a leading voice-over video application in China, 'Kuaipai' was previously known for features such as script readers and intelligent video mixing. This upgrade will further enhance its AI creation capabilities.

Rivaling Claude 4.5! Silicon-based Flow Launches High-speed Version GLM-5, Domestic Large Model Secures Fourth Place Globally

The domestic large model GLM-5 achieved significant breakthroughs in early 2026, ranking fourth in the authoritative list Artificial Analysis globally after being open-sourced, with a score comparable to Claude Opus4.5. Its core innovations include: expanding parameter scale to 744B and pre-training data reaching 28.5T; integrating DeepSeek sparse attention mechanism, which maintains long-text understanding capabilities while reducing deployment costs; and excelling in programming and engineering fields.

Xiaomi Open Sources First-generation Robot VLA Large Model, Breaking the Bottleneck of Physical Intelligence Latency

Xiaomi opensources the 4.7 billion parameter robot model Xiaomi-Robotics-0, which adopts a MoT hybrid architecture. By enabling collaboration between the 'brain' and 'cerebellum', it achieves real-time inference on consumer-grade GPUs, solving the problem of sluggish actions caused by inference latency in existing VLA models, thus improving the efficiency and generalization ability of robot control.

Dual Evolution of Intelligent Driving and Cockpit! AVATR.OS 5.0 Officially Released: MoLA Large Model on Board, First Direct Upgrade to Huawei ADS 4.1

AVATR released the AVATR.OS 5.0.0 system for all vehicle models on February 11, 2026. The core of this major update is the deep integration of AI large model capabilities and upgrading to Huawei's latest intelligent driving system. The most notable highlight is the official launch of the MoLA large model assistant, which has significantly enhanced semantic understanding capabilities, supports word composition, character splitting, and knowledge correction, allowing more accurate understanding of user commands.