Next-generation large language model optimized for coding and reasoning tasks.
Expected to launch as GPT-5.2 or GPT-5.5 in Q1 2026.
OpenAI is developing a new large language model codenamed "Garlic" to compete with Google's Gemini 3 and Anthropic's Opus 4.5. The model has shown exceptional performance in internal benchmarks for coding and reasoning tasks.
According to reports from The Information (December 2025), Chief Research Officer Mark Chen shared that Garlic has outperformed competing models in key areas. A notable technical breakthrough is the solution to pretraining challenges, allowing smaller models to be injected with more knowledge while improving efficiency.
Key indicators and announcements about the Garlic model development.
OpenAI has declared 'Code Red' to address competitive pressure from Google and Anthropic.
Internal tests show Garlic outperforming Gemini 3 and Opus 4.5 in coding and reasoning.
Novel approach allows smaller models to contain more knowledge with improved efficiency.
Expected public release in Q1 2026 as either GPT-5.2 or GPT-5.5.
Competitive response to Google's Gemini surge and rising AI industry competition.
Focus on ChatGPT competitiveness and core model quality improvements.
Track the development progress of the Garlic model.
Mark Chen shares Garlic's strong benchmark results with OpenAI team.
Sam Altman declares Code Red, refocusing company on core model quality.
Continued pretraining and evaluation against competitors.
Expected launch as GPT-5.2 or GPT-5.5.
How Garlic compares to other frontier models.
| Dimension | Garlic | Gemini 3 | Claude 4·5 | GPT-5·1 |
|---|---|---|---|---|
| Coding Performance | Excellent | Very Good | Very Good | Good |
| Reasoning Ability | Excellent | Very Good | Excellent | Good |
| Training Efficiency | High | Medium | Medium | Medium |
| Expected Release | Q1 2026 | Released | Released | Released |
Projected performance metrics based on internal evaluations.
| Benchmark | Expected | Actual |
|---|---|---|
| SWE-bench | >70% | TBD |
| MMMU-Pro | >80% | TBD |
| Humanity's Last Exam | Top Tier | TBD |
| HumanEval | >95% | TBD |
| MATH | >90% | TBD |
| GPQA Diamond | >75% | TBD |
Data based on internal evaluations and industry reports. Official benchmarks pending public release.
What makes the Garlic model stand out from the competition.
Novel approach that injects more knowledge into smaller models while maintaining efficiency.
Exceptional performance in code generation, debugging, and understanding large codebases.
Advanced logical reasoning and problem-solving capabilities for complex tasks.
Trained on smaller datasets while maintaining comparable sophistication to larger models.
Designed to handle autonomous agent workflows and multi-step operations.
Developed with built-in safety measures and responsible AI principles.
How different user groups can leverage the Garlic model.
Prepare your organization for the Q1 2026 release.
Establish benchmarks and success metrics for your specific use cases.
Allocate resources for API costs and integration development.
Define acceptable response times for your applications.
Ensure compliance with data handling and privacy policies.
Design API integration patterns and fallback strategies.
Upskill team on prompt engineering and model capabilities.
Common questions about OpenAI's Garlic model.
Have more questions? Follow the latest updates from OpenAI.
Verified sources for the information presented on this page.