XNews

GPT-5.6 Model Rollout and User Reactions

GPT-5.6 Model Rollout and User Reactions

OpenAI rolled out the GPT‑5.6 model family, introducing multiple variants and new user controls. • Free and Go users gain unlimited access to the new default model GPT‑5.6 Luna for text chat. • GPT‑5.6 Sol powers Plus and Pro users, tuned to give more direct, concise answers with less redundant pleasantries. • Factual error rate on complex finance, medical, and legal tests dropped by 68% compared to GPT‑5.5. • A thinking‑depth slider lets users choose quick responses or deeper reasoning for coding/research tasks; free users get a “Think” button. • Youth safety features forbid emotional/romantic role‑play, discourage use as a human relationship replacement, and strengthen filtering for self‑harm, eating disorders, and dangerous content, with prompts to seek trusted adult help. • Perplexity tested GPT‑5.6 Terra extensively and found it capable and cost‑effective; it becomes the default model for all sub‑agents and an orchestrator option for Perplexity Computer users. • A user noted that GPT‑5.6 answers tend to be long and language quality sometimes falls below Claude’s dense model output, but the free Luna model is considered usable with limited thinking access.

Key insights

OpenAI targets factual reliability and safety with GPT‑5.6

The update reduces factual errors by 68% over GPT‑5.5 in high‑stakes domains such as finance, medicine, and law. New youth safety fine‑tuning explicitly prohibits romantic role‑play and actively discourages using AI as a substitute for human relationships, while strengthening filters for self‑harm and dangerous activities. A thinking‑depth slider and a Think button for free users give more control over reasoning time.

Perplexity adopts GPT‑5.6 Terra for cost‑effective multi‑agent orchestration

Perplexity CEO Arav Srinivas stated that after extensive testing, GPT‑5.6 Terra proved both capable and cost‑effective, leading to its adoption as the default model for all sub‑agents inside Perplexity Computer and an orchestrator option for all users. The integration highlights the model’s suitability for complex, multi‑model workflows in a commercial product.

Users criticize GPT‑5.6’s verbosity and language quality compared to Claude

A user observed that GPT‑5.6 often provides very long responses with lower language quality, and expressed a preference for Claude’s dense model for question‑answering. The criticism points to a persistent gap in prose style, despite the model’s other improvements.

Free GPT‑5.6 Luna seen as adequate for everyday use

Another user noted that free and Go users now get unlimited GPT‑5.6 Luna, and the free tier is already usable with limited access to the thinking feature. This suggests the baseline model is sufficient for many casual tasks, lowering the barrier to advanced AI assistance.

Source posts