GPT-5.3 Garlic: What OpenAI officially released and what verified leaks say
Next step
Open the bot or continue inside this section.
Article -> plan in AI
Paste this article URL into any AI and get an implementation plan for your project.
Read this article: https://vibecode.morecil.ru/en/%D0%BD%D0%BE%D0%B2%D0%BE%D1%81%D1%82%D0%B8/gpt5.3-garlic/
Work in my current project context.
Create an implementation plan for this stack:
1) what to change
2) which files to edit
3) risks and typical mistakes
4) how to verify everything works
If there are options, provide "quick" and "production-ready". How to use
- Copy this prompt and send it to your AI chat.
- Attach your project or open the repository folder in the AI tool.
- Ask for file-level changes, risks, and a quick verification checklist.
OpenAI continues the rapid cycle of updates to the GPT-5 line. After the flagship GPT-5 (August 2025) and GPT-5.2 (December 2025) in February 2026, the company introduced the first models under the number 5.3 - specialized GPT-5.3-Codex and GPT-5.3-Codex-Spark.
These releases are fully confirmed by official announcements on openai.com. The codename "Garlic"( ) is actively discussed in the community and is associated with a possible general release of GPT-5.3, but OpenAI has not yet confirmed it. Below are only verified facts from official sources + clearly marked rumors from reliable leaks (Dan McAteer et al.).
Official GPT-5 lineup for today
- GPT-5 – August 2025: basic model of new generation.
- GPT-5.2 – December 11, 2025: focus on professional work, long context (up to 256k high-precision tokens), agent scenarios, mathematics and science. New records for GDPval (70.9%), GPQA Diamond (92.4%), AIME 2025 (100%), etc.
- GPT-5.3-Codex – ** February 5, 2026**: the first model of the 5.3 family.
- GPT-5.3-Codex-Spark — February 12, 2026**: lightweight ultra-fast version.
GPT-5.3-Codex: the most powerful agentic coding model to date
** Release date:** February 5, 2026.
** Key features (directly from OpenAI announcement):**
- Combines the best coding abilities of GPT-5.2-Codex and strong reasoning + professional knowledge of GPT-5.2 in one model.
- 25% faster* than previous versions thanks to infrastructure optimizations (NVIDIA GB200 NVL72).
- Full agentic workflow: The model can conduct long multi-step tasks (research, tool use, execution) and you direct it in real time without losing context.
- For the first time, the **sama model participated in its creation: the Codex team used early versions to debug training, deploy and analyze tests.
What does he do in practice
- Builds complex web applications and games in a matter of days (with aesthetics and understanding of user intent).
- Full development cycle: from PRD and code to demoment, monitoring and metrics.
- Professional tasks: presentations, data analysis, user research.
- Cybersecurity: The first OpenAI model to be rated “High Capacity” by the Preparedness Framework is looking for vulnerabilities, and there is a “Trusted Access for Cyber” pilot.
** Official benchmarks:**
- SWE-Bench Pro: 56.8% (GPT-5.2-Codex 56.4%)
- Terminal-Bench 2.0: 77.3% (up from 64%)
- OSWorld-Verified: 64.7% (up from 38.2%)
- GDPval: 70.9% (at GPT-5.2)
- Cybersecurity CTF: 77.6%
- SWE-Lancer IC Diamond: 81.4%
Available now to all paid users of ChatGPT through Codex app, CLI, IDE plugins and the web. API rollout coming.
GPT-5.3-Codex-Spark: Real-time coding on steroids
** Release date:** February 12, 2026 (research preview).
- 15x faster than generation.
- 128k context.
- More than 1,000 tokens per second on Cerebras equipment (partnership announced in January 2026).
- Ideal for interactive coding on the fly.
"Garlic" is the code name for the next step (only rumors, but very persistent)
OpenAI has not officially confirmed the Garlic name, nor has it announced a generic (non-Codex) GPT-5.3 model. However:
- Numerous leaks (including posts by Dan McAteer @daniel mac8, often cited as a reliable source) are referred to as GPT-5.3 "Garlic".
- A/B tests of the new version are already available in the ChatGPT app (you’re giving feedback on a new version of ChatGPT).
- Expected leak release date: 26 February 2026 (tomorrow).
What the leaks promise (not confirmed by OpenAI):
- 400k context tokens + 128k output.
- High-density training: not just more parameters, but smarter and more efficient through data curation and compression.
- Native agent capabilities + self-verification.
- SimpleBench: 83.7% (exceeds human baseline; previous models were significantly lower).
- The overall jump in quality is compared to the transition of GPT-3 → GPT-4.
OpenAI purportedly declared an internal "Code Red" due to the pressure of Gemini 3 and Claude Opus 4.5/4.6 and focused on efficiency and reasoning, not just on scale.
Why this is important for developers
GPT-5.3-Codex is already turning Codex into a full-fledged “colleague engineer” who can lead the project for weeks. If the rumors about Garlic are confirmed, tomorrow we will get a universal model that:
- he has a huge codebase in his head,
- independently plan and perform tasks,
- it works much more efficiently and cheaper.
For freelancers, startups and enterprise teams, this is the next step in getting one senior developer + GPT-5.3 to do the work of a whole team.
Result (as of February 25, 2026): Officially we have the most powerful coding agents GPT-5.3-Codex and Spark. "Garlic" remains the industry's hottest rumor with the release of "tomorrow." As soon as OpenAI announces something, we will immediately update the article.
Follow the updates on vibecode.morecil.ru - we keep abreast of all significant OpenAI releases.