OpenAI has discontinued its standalone Codex coding model, consolidating its capabilities into the upcoming GPT-5.5. The shift marks another step in the company's move toward unified language models.
OpenAI has officially ended support for Codex, its dedicated code generation model, folding its functionality into GPT-5.5.
Codex, released in 2021, was specifically trained to understand and generate code across dozens of programming languages. It powered GitHub Copilot's initial versions and attracted developer interest as an AI-first coding tool.
The decision to retire the standalone model reflects a broader industry trend toward consolidated AI systems. Rather than maintaining separate models for different tasks, companies increasingly build general-purpose models capable of handling multiple domains.
GPT-5.5, OpenAI's next-generation model, will incorporate Codex's coding capabilities alongside its existing language understanding. The company expects the unified approach to deliver better performance by leveraging improved training techniques and larger datasets.
Developers currently using Codex via API will need to migrate to GPT-5.5 before the discontinuation date. OpenAI has not specified an exact sunset timeline but indicated a transition period will be provided.
This is the second time OpenAI has consolidated Codex. The company previously merged its original GPT-4 Turbo model with standard GPT-4, moving toward a cleaner product lineup.
The consolidation raises questions about specialized model development in AI. While unified models offer advantages in maintenance and resource allocation, some developers argue that dedicated models better serve specific use cases.
OpenAI's approach aligns with industry momentum. Anthropic and Google have similarly opted for multipurpose models rather than task-specific variants, suggesting the market may be moving away from fragmented AI tooling.
GPT-5.5 availability details and pricing for the new unified model remain under wraps. OpenAI is expected to announce full specifications and migration guidelines in coming weeks.
Multiverse Computing has unveiled Quasar 438B, positioning it as Europe's leading AI model. The release marks a significant step in the region's efforts to develop competitive large language models.
Google has launched Gemini 3.8 Flash just three weeks after releasing version 3.7 Flash. The new model is available at introductory rates of $0.75 per million input tokens and $3.75 per million output tokens through December 31.
Three websites created over 215,000 "best software" pages targeting AI queries, with Perplexity AI citing these pages as sources. The manufactured content raises questions about AI training data quality and source reliability.
Mistral AI now uses user inputs and outputs for model training by default across its platform, with opt-out available only to enterprise customers. The policy applies to all free and paid tiers except the enterprise tier.