GPT-6 Sol/Luna and Claude Opus 5.5: Costs Decrease

Key Takeaways

1OpenAI and Anthropic launch new models focused on speed and cost reduction

2Claude Opus 5.5 shows a 40% lower cost compared to Opus 5 and increased usage limits

3GPT-6 Sol and Luna aim to make the power of Astra more accessible, with higher usage ceilings

Why it matters : These developments enhance the competitiveness of AI offerings for different user profiles, while emphasizing cost control and response quality.

OpenAI and Anthropic are rolling out new versions of their flagship models, focusing on speed, cost reduction, and increased usage limits. Both companies also claim advancements in alignment and factual accuracy, with differentiated access based on their offerings.

Increased Usage Limits and Expanded Access Based on Plans

OpenAI is launching GPT-6 Sol and GPT-6 Luna for ChatGPT Work and Codex, while GPT-6 Luna remains accessible to free users and lower-tier subscribers. Anthropic is raising usage limits by 5 hours on its Pro, Max, and Team plans.

Announced Cost Reductions and Efficiency

The new models emphasize faster execution and lower costs compared to previous generations. Anthropic states that Claude Opus 5.5 matches the capabilities of Fable 5.1 while reducing costs by nearly 40% compared to Opus 5, and notes that the new model requires less computational power. According to internal tests with default parameters, Anthropic also estimates a 40% cost reduction on typical workloads.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Claims of Factual Accuracy, Alignment, and Safeguards

OpenAI claims to have strengthened factual accuracy in its latest ChatGPT models, stating they make less than half the errors compared to previous versions. Anthropic presents Opus 5.5 as its most advanced model in terms of alignment, specifying that it incorporates the same safeguards as its high-end models, with the option to revert to an older model for certain cybersecurity or scientific tasks. Both companies also announce more natural communication, with Anthropic noting improvements in writing and instruction-following based on its tests and user feedback.

Positioning and Release Schedule for GPT-6 and Astra

GPT-6 Sol and GPT-6 Luna succeed the 5.6 versions released in July and arrive shortly after the launch of ChatGPT Astra in early September. OpenAI designed Sol and Luna to make the power of Astra more accessible to lower budgets while increasing usage limits for businesses and developers, maintaining similar training methods and improvements in coding and computer usage. OpenAI continues to present Astra as the "best model in the world" for computing use. These announcements come as new models for Claude and ChatGPT are now available and were launched on Tuesday.