Codex Mini by OpenAI

Key takeaways

• Codex Mini is a Transformer-based model primarily fine-tuned for coding assistance within the Codex CLI.
• It excels at elementary science, symbolic manipulation, algebra, and geometry problems, achieving 98.3% accuracy on the AI2 Reasoning Challenge - Easy dataset.
• The model struggles with broad reasoning, complex calculations, long prompts, intricate arithmetic, common sense, and identifying relevant relationships.
• Codex Mini scored 0% accuracy on the Logical Fallacy Detection Benchmark, consistently returning empty responses, indicating a fundamental lack of critical thinking and complex reasoning.
• Its performance on the Logical Fallacy Detection Benchmark highlights its inability to perform multi-class classification or explanation generation, failing to produce any meaningful output.

Technical specifications

Context length

200k tokens

Max output

—

Architecture

Transformer

Modalities

text+image->text

Parameters

~13B

Open weights

No

Licence

Proprietary

Regions

USA

Vendor

OpenAI

Provider model key

openai/codex-mini