OpenAI sidesteps Nvidia with unusually fast coding model on plate-sized chips

Summary

OpenAI released GPT-5.3-Codex-Spark, a coding model running on Cerebras chips, achieving 1,000 tokens per second. This model is faster than its predecessor and aims to reduce OpenAI's reliance on Nvidia hardware.

Key Takeaways

1. OpenAI released GPT-5.3-Codex-Spark, its first production AI model to run on non-Nvidia hardware, using Cerebras chips.
2. Codex-Spark achieves code delivery at over 1,000 tokens per second, reportedly 15 times faster than its predecessor.
3. OpenAI is diversifying away from Nvidia by partnering with AMD, Amazon, and designing its own custom AI chip.

Statement Breakdown

• Claimed Facts: 70% of statements the article presents as facts
• Opinions: 20% of statements classified as editorial or subjective
• Claims: 10% of statements surfaced for additional reader evaluation

Credibility & Bias Reasoning

Credibility assessment: The article is published on Ars Technica, a reputable tech news source. It cites specific benchmarks and statements from OpenAI and Cerebras. The author also acknowledges limitations and potential drawbacks, enhancing credibility.

Bias assessment: Technological Advancement Focus. The article emphasizes the technological advancements and competition in the AI coding agent space. It frames the news through the lens of innovation and market dynamics, with a slight inclination towards celebrating progress.

Note: While the article presents factual information, some claims rely on company-provided data. Consider independent validation when evaluating performance metrics.

Credibility flag: Mostly Reliable

Claimed Facts (7)

• This is a verifiable statement of fact regarding the release of a new AI model.
• This is a quantifiable performance metric.
• This is a direct quote from a named individual at OpenAI.
• This is a factual statement about the availability and access to the model.
• This is a statement about OpenAI's strategic direction.
• These are specific, verifiable business deals.
• Describes the hardware used by the model.

Opinions (5)

• This is a subjective assessment of the impact of AI coding agents.
• This is an interpretation of the competitive landscape.
• This is a subjective assessment of the importance of the hardware.
• This is a subjective comparison using figurative language.
• This is a subjective warning.

Claims (5)

• This claim relies on OpenAI's own reporting without independent validation.
• Highlights the lack of independent verification, making the claim dubious.
• This claim is based on a report, but the original source of the dissatisfaction is not directly cited or verified.
• The 'code red' memo is anecdotal and lacks concrete evidence.
• While presented as independent, the specific methodology and reliability of 'Artificial Analysis' are not detailed.

Key Sources

• Benj Edwards — Author
• Sachin Katti — head of compute at OpenAI
• OpenAI — Company
• Reuters — News organization
• Artificial Analysis — Independent Benchmarking