OpenAI's GPT-5 arrives as Claude Opus 4.1 counters with coding upgrades

OpenAI has finally shipped GPT-5, a unified AI system that dynamically chooses between fast responses and extended reasoning based on query complexity. The model delivers state-of-the-art results: 94.6% on AIME 2025 math, 74.9% on SWE-bench Verified coding, and a 45% reduction in hallucinations versus GPT-4o. GPT-5 is available immediately to all ChatGPT users, with Pro subscribers getting GPT-5 Pro for deeper reasoning on complex tasks. API pricing starts at $1.25 per million input tokens and $10 per million output, with Mini and Nano tiers dropping to as low as $0.05/$0.40 — making advanced AI remarkably cheap. OpenAI also highlighted "safe completions" that reduce refusals while improving safety.

Anthropic fired back with Claude Opus 4.1, its flagship model tuned for complex coding, autonomous research, and agentic workflows. Opus 4.1 introduces hybrid reasoning with user-controlled "thinking budgets" for cost optimization, and claims industry-leading SWE-bench performance. It's available via API, Amazon Bedrock, and Vertex AI at $15/$75 per million tokens, with prompt caching and batch discounts up to 90%. Elsewhere, Google DeepMind previewed Genie 3, a world model generating interactive 3D environments at 24fps 720p from text prompts — a breakthrough for gaming and robotics simulation. ElevenLabs launched Eleven Music, a commercial text-to-music service with licensing from Merlin and Kobalt at $0.50 per generated minute. Cogito open-sourced four V2 reasoning models, including a 671B MoE that matches DeepSeek's performance.