Releases, benchmarks and capability jumps from OpenAI, Anthropic, Google DeepMind, Meta and xAI.
Frontier models are the largest, most capable general-purpose AI systems at the edge of what is technically possible — today that means GPT-class models from OpenAI, Anthropic's Claude, Google DeepMind's Gemini, Meta's Llama and xAI's Grok. This section tracks every major release, benchmark result and capability shift across the leading labs, with sourced analysis of what each change means for the people who build on these models.
The premium tier targets agentic coding workloads on the GLM-5.3-Flash base with expanded speed, context, and input modalities while maintaining separate access from standard subscriptions.
Alibaba's Qwen team released the model on September 18, 2026, offering 1M-token context and audio-video inputs through pay-as-you-go services that compete directly with Gemini and DeepSeek options on platforms including OpenModels.
Shanghai Artificial Intelligence Laboratory open sources a 744B-parameter agentic model on GLM-5.2 weights under MIT license that leads on several agent benchmarks and supports verifiable tool use.
The September 17, 2026, release combines GPT-6 Astra with a 230 million URL legal index and custom tools, initially available to selected firms via trusted access in ChatGPT and Codex.
The September 2026 launch introduces native support for text, image, audio and video inputs alongside a 1M-token context window, cost reductions exceeding 93 percent in some cases, and open-source tools that extend the model into real-world productivity and creative execution.
The 8-billion parameter model fine-tuned from Hermes-3-Llama-3.1 supports specialized tasks in Solana perps research and agent development while remaining distinct from live execution systems.
The September 2026 release of antigravity-preview-09-2026 for Gemini managed agents introduces file management tools, secure credential registration, and caching enhancements that reduce costs on complex workflows while maintaining existing pricing.
The open-weight agentic model built on GLM-5.2 leads specific benchmarks and targets research and engineering workflows through a verifiable training approach.
Released on September 15, 2026, the new Gemini audio models support background tool use and parallel reasoning, achieving leading positions on speech-to-speech quality and agentic benchmarks while integrating with Search and Workspace products.
Google releases Gemini 3.8 Live for immediate developer use via API and AI Studio, powering Search Live with camera input, background lookups, and 97-language support while the Extended Thinking variant adds multi-step reasoning.
Human preference votes place the model ahead of competitors while Anthropic systems maintain a strong presence across multiple top ranks in the latest snapshot.
An audit alleges that financial flows from Anthropic's equity to safety evaluators create a self-reinforcing regulatory capture mechanism, with Dustin Moskovitz's donated stake reaching a potential $7.7 billion valuation.
The September 14 2026 preview release highlights verifiable experience pipelines and draws on human-AI collaboration data from 769 task records to target research and engineering workflows.
The InternLM team open-sources a large-scale model that scales pre-training, reinforcement learning across scientific domains, and agent environments to support long-horizon tasks.
The new model activates far fewer parameters than its size suggests while adding native vision and a 1 million token context at sharply reduced API rates.
The September 11, 2026 releases demonstrate that orchestration of open models can deliver top benchmark results at lower costs while avoiding dependence on closed frontier systems such as GPT-6 Astra and Claude Fable 5.1.
The 552B-parameter release ships with native vision, MIT weights, 1M context and Flash-tier pricing while routing legacy Pro traffic to the new endpoint from mid-September.
The September 10, 2026 report covers misuse disrupted from December 2025 to August 2026, including seven China-based labs targeting Claude through distillation and a fourth cybersecurity incident with an early Claude Opus 4.6 version.
The new model collapses separate ASR, TTS and LLM components into one architecture that supports simultaneous listening and speaking, with API access, evals and pricing now public alongside expanded ChatGPT Voice availability.
The 552B-parameter model introduces CED architecture, Engram memory and DSpark decoding to cut KV cache costs fourfold while outperforming its V4 Pro predecessor on benchmarks and triggering its retirement.
A frontier model is a general-purpose AI system trained at the largest compute scales — models like GPT-class, Claude, Gemini, Llama and Grok — whose capabilities exceed prior systems on broad benchmarks.
Major labs ship a flagship roughly every 6–12 months, with point upgrades in between. This hub updates on every confirmed release.
Each article here cites the primary benchmark source — the lab model card or an independent eval — so you can verify the numbers directly.