Topic Hub · AI
Frontier Models
Frontier models define the ceiling of what AI can do today — and shape investment, regulation, and competitive strategy tomorrow. Credence Wire tracks releases, capabilities, safety evaluations, and the economics of training at scale.
Edited by Rahul Subramaniam, Priya Mehta, Sara Chen
Updated June 7, 2026
Latest Frontier Models Stories
Latest Frontier Models Coverage
OpenAI Signs Multi-Year Licensing Deals With Five Global News Publishers
Agreements with Axel Springer, Le Monde, Prisa, Nikkei, and The Associated Press grant training and display rights in exchange for revenue shares on referred subscriptions.
Apple Introduces On-Device Agents That Run Without Cloud Access at WWDC
The company said its new AgentKit framework executes multi-step tasks locally on iPhone and Mac, addressing enterprise concerns about data leaving controlled environments.
Anthropic Closes $65 Billion Round at $965 Billion Valuation, Leapfrogging OpenAI
The San Francisco AI lab's Series H makes it the world's most valuable private AI company, with Samsung, SK Hynix, and Micron joining hyperscaler backers ahead of a widely expected IPO.
Claude Opus 4.8 Ships as Anthropic Prepares Mythos for Broader Release
The upgraded flagship model is four times less likely to let code flaws pass unremarked, while Project Glasswing partners have used Mythos Preview to surface more than 10,000 critical vulnerabilities.
Cognition Raises $1 Billion at $25 Billion Valuation as Devin Enterprise Revenue Surges
The autonomous coding startup more than doubled its valuation in eight months, betting that independent AI software engineers can survive the model makers' push into the same market.
UN General Assembly Adopts Global AI Governance Framework
The non-binding pact commits signatories to share safety evaluations and restrict autonomous weapons development.
Explore Related Topics
Generative AI
Foundation models, creative tools, and the commercial transformation of generative systems.
AI Safety
Research, policy, and industry efforts to align powerful AI systems with human values.
AI Infrastructure
Data centers, GPUs, cloud platforms, and the physical layer of the AI economy.
Frontier Models Tracker
Live editorial updates
Frontier Models Membership
Premium analysis and briefings
Frequently Asked Questions
What counts as a frontier model?+
Frontier models are the most capable AI systems at a given point in time, typically trained on the largest compute budgets and achieving state-of-the-art results across a broad range of benchmarks. Current examples include GPT-4o, Claude 3.5, and Gemini Ultra.
How are frontier models evaluated?+
Evaluations include academic benchmarks (MMLU, HumanEval), red-teaming for safety risks, capability elicitation tests, and increasingly, third-party audits. No single benchmark captures full capability.
What is the compute threshold for frontier models?+
Policymakers and researchers often use training compute as a proxy — the EU AI Act and US executive orders reference thresholds around 10^26 FLOPs, though the frontier shifts as hardware and efficiency improve.
Credence Wire Membership
Get deeper frontier models intelligence
Get premium analysis, member-only briefings, and ad-free access across all topic hubs.