What's New

GPT-6 Astra (gpt-6-astra) is now available in the model picker and the AI workflow designer — OpenAI's frontier GPT-6 model, a step-change over the GPT-5.6 family (Sol/Terra/Luna) on deep reasoning, agentic tool use, and long-horizon work, with a 2M token context window. Select it under the OpenAI provider anywhere a model can be chosen. GPT-5.6 Terra remains the recommended production default for OpenAI.

Zhipu AI's latest GLM release is now available via the Ollama Cloud provider (glm-5.3). It supersedes GLM-5 with major gains in coding and agentic-task performance from scaled post-training, and a 1M token context window.

Four new models landed in the model picker:

• Claude Fable 5.1 (claude-fable-5-1) — Anthropic's most capable widely-released model, succeeding Claude Fable 5 in the same tier and at the same price.
• Gemini 3.8 Flash (gemini-3.8-flash) — Google's most intelligent workhorse yet, with major gains on software engineering and agentic tasks over 3.7 Flash.
• Gemini 3.7 Flash (gemini-3.7-flash) — the prior-generation Flash release, still fully supported for lower-token-usage workloads.
• Grok 4.6 (grok-4.6) — xAI's new flagship, succeeding Grok 4.5 with configurable reasoning effort up to xhigh.

Claude Opus 5 (claude-opus-5) is now available — the new flagship of the Opus tier, a step-change over Opus 4.8 on deep reasoning, agentic coding, and long-horizon work, at the same $5/$25-per-million-token pricing as its predecessor. Now the recommended Opus-tier pick on the platform.

Added Claude Fable 5, Claude Opus 4.8, and Claude Sonnet 5 (Anthropic); GPT-5.6 Sol/Terra/Luna (OpenAI); Gemini 3.6 Flash and Gemini 3.5 Flash-Lite (Google); Grok 4.5 (xAI); Kimi K3/K2.6 (Ollama Cloud); and Llama 4, Qwen 3.6, Mistral Small 4 (HuggingFace). Also refreshed each provider's recommended default model and fixed several default-model IDs that no longer matched any model in the catalog.

Organizations can now export a full, tamper-evident audit trail of account activity from Settings, with a dedicated log viewer and CSV export — built for compliance reviews.

Fixed a set of service-role API routes that weren't fully enforcing role-based access control, ensuring permission checks are applied consistently across the platform.

Marketplace templates now track how often they're copied, surfacing the most-used community workflows.

Google integration reconnect-needed emails are now suppressed for 3 days after the first notification, instead of firing on every check.

Projects can now be renamed or deleted directly from their card menu, without opening the project first.