OpenAI Releases GPT-5.6 Series
On July 10, OpenAI officially launched the GPT-5.6 series alongside ChatGPT Work for long-running tasks, and merged Codex into the new ChatGPT desktop app.
Three Models
The GPT-5.6 series comes in three tiers:
- Sol (flagship) — $5/M tokens input, $30/M tokens output
- Terra (balanced) — $2.5/M tokens input, $15/M tokens output
- Luna (lightweight) — $1/M tokens input, $6/M tokens output
All three models are available immediately to ChatGPT, Codex, and API users.
Ultra Mode
The new series introduces Max reasoning intensity and Ultra mode. In Ultra mode, the model coordinates 4 parallel agents for complex tasks. According to OpenAI, GPT-5.6 Sol scores 88.8% on Terminal-Bench 2.1, improving to 91.9% with Ultra mode enabled.
ChatGPT Work
ChatGPT Work handles long-running workflows, pulling information from apps and files to produce research analysis, documents, spreadsheets, presentations, and websites.
Desktop Integration
The original Codex desktop app has been upgraded to the new ChatGPT desktop app, offering Chat, Work, and Codex in one place. The old ChatGPT desktop app is now called ChatGPT Classic. OpenAI also announced ChatGPT Atlas browser will be discontinued on August 9, with its browser agent capabilities moving to ChatGPT and Codex.
SpaceXAI Launches Grok 4.5
On July 9, SpaceXAI released Grok 4.5, its first model specifically trained for programming and agent tasks, co-developed with Cursor. Musk described it as an "Opus-level model."
Grok 4.5 is designed for real engineering scenarios — large codebases, cross-repository work, multi-tool integration, and hundreds of skills. It was trained on tens of thousands of NVIDIA GB300 GPUs, with deduplication, quality scoring, and domain focusing during data processing, and reinforcement learning covering hundreds of thousands of tasks.
Beyond code, Grok 4.5 handles office tasks: building Excel models in Grok Build, creating presentations in PowerPoint, and writing documents in Word. It can generate end-to-end apps from simple prompts, supporting Rust, C, and C++.
ByteDance Releases Seedream 5.0 Pro
On July 8, ByteDance launched Seedream 5.0 Pro, a multimodal image generation model. Compared to its predecessor, it improves text-image matching, composition, text rendering, and visual quality.
Seedream 5.0 Pro offers better spatial awareness and semantic understanding, supporting point-select, circle-select, sketch rendering, color and material replacement, layer separation, and multi-image composition. Lighting, material, and skin texture rendering have been improved.
The model supports text rendering in over a dozen languages. It's now available on Volcano Ark experience center, with plans to integrate into Doubao and Jimeng.