OpenAI Introduces GPT-6.1 Sol with 2M Token Context, Launches Model Marketplace

OpenAI announced GPT-6.1 Sol, its newest flagship model, at its annual DevDay conference. This model offers expanded capabilities, including a 2M token context window. Sol also features a 4.5 times increase in rate limits compared to its predecessor. OpenAI stated that Sol processes 200,000 tokens per second for input and 600,000 tokens per second for output. It integrates advanced multimodal reasoning across text, audio, and visual inputs.

GPT-6.1 Sol demonstrates improved performance on various benchmarks. OpenAI reported that the model achieved a 95.1% score on the MMLU benchmark. On the GPQA Diamond benchmark, Sol scored 94.7%. It also attained a 93.8% score on the MATH benchmark. The model’s enhanced performance extends to complex reasoning and code generation tasks. Sol features a 2.5-second audio response time for conversational applications. OpenAI has released GPT-6.1 Sol for immediate developer access.

The company also introduced an AI Model Marketplace. This platform allows developers to access and deploy custom models. The marketplace supports fine-tuned models and bespoke architectures. OpenAI emphasised a revenue-sharing model for marketplace contributors. This initiative aims to foster a broader AI development community. The marketplace includes models from various third-party developers and research institutions.

Further announcements included a new Assistants API. This API simplifies the creation of AI agents. It incorporates capabilities such as code interpretation and knowledge retrieval. Developers can build assistants with persistent threads and custom tool use. The Assistants API supports stateful interactions over extended periods. It offers a structured approach to integrating sophisticated AI logic into applications.

OpenAI also previewed improvements to its DALL-E model. DALL-E 4 will offer enhanced image generation quality and control. The new version introduces features like inpainting and outpainting. It also provides greater semantic understanding for prompts. OpenAI plans to roll out DALL-E 4 to developers in the coming months.

Updates to the GPT-4 Turbo model were also detailed. GPT-4 Turbo now includes a 512k token context window. Its knowledge cut-off extends to June 2024. This version offers improved instruction following and JSON mode support. It also features reproducible outputs, essential for consistent application behaviour. OpenAI described GPT-4 Turbo as a more cost-effective option for developers.

Pricing for GPT-6.1 Sol was announced. Input tokens are priced at $10 per million tokens. Output tokens are priced at $30 per million tokens. OpenAI positioned Sol as a premium offering for advanced AI applications. The new pricing structure reflects the model's increased capabilities and efficiency. This development aims to balance performance with economic accessibility for developers.

These announcements reinforce OpenAI's strategy. The company aims to provide a comprehensive AI development stack. It covers foundational models to deployment platforms. The focus remains on empowering developers. Will these new tools accelerate the commoditisation of AI applications?