GPT-5.3-Codex: terminal-Bench 2.0 is documented at 77.3%, rather than 78.4%
Performance discrepancies in Terminal-Bench 2.0 reveal transparency gaps in OpenAI's latest documentation.
Terminal-Bench 2.0 is documented at 77.3%, rather than 78.4% — verified by the OnAirToday AI Progress Tracker against OpenAI's published documentation.
OnAirToday
Newsroom
The OnAirToday AI Progress Tracker recorded a verified change to OpenAI's GPT-5.3-Codex (version gpt-5.3-codex): Terminal-Bench 2.0 is documented at 77.3%, rather than 78.4%.
GPT-5.3-Codex is OpenAI’s agentic coding model for software development and longer-running tasks involving tools and computers. It combines coding and general reasoning capabilities, with additional safeguards reflecting OpenAI’s High cybersecurity capability classification.
What changed in this review:
- SWE-bench Verified: Not yet confirmed by public reporting.
- Added documented coding, tool-use and interactive steering capabilities.
- Added OpenAI’s High cybersecurity capability classification and associated safety concerns.
- Flaws and limitations updated
- Risks and threats updated
Teams building on GPT-5.3-Codex should confirm limits, pricing and benchmark coverage against the vendor's own documentation before relying on these figures.
The full profile — capabilities, benchmark citations, known flaws and risk guidance — is on the tracker at /ai-models/gpt-5-3-codex.
References:
- https://openai.com/index/introducing-gpt-5-3-codex/
- https://openai.com/index/gpt-5-3-codex-system-card/
- https://developers.openai.com/api/docs/models/gpt-5.3-codex
Generated insight by OnAirToday AI
The discrepancy in Terminal-Bench 2.0 performance metrics highlights the critical need for independent verification in the rapidly evolving landscape of LLM benchmarks. Accurate reporting is essential for developers to correctly assess the real-world capabilities and reliability of cutting-edge coding models.
Comments
Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.