The latest moves
in machine intelligence.

Keeps the Flash endpoint aligned with DeepSeek’s newest model revision.

Extends Mistral into native forecasting for operational and scientific data.

Advances the live audio line with a newer safety-reviewed deployment.

Adds a distinct live-audio personality variant with the 1.1 safety controls.

Moves xAI’s voice stack to a newer generation in the official release notes.

Extends the Claude 5.5 family to real-time and high-volume workloads at sharply lower prices.

Mistral’s largest model so far uses 49B active parameters and is intended for an open-weight release.

Teams can index for quality and query for speed without rebuilding their corpus.

Materially improves the cost-performance frontier for capable agents.

Runs more than 30% faster than Sonnet 5 and costs up to 30% less for most work.

Brings GPT-6 intelligence to everyday agents while halving API prices from GPT-5.6 Sol.

Extends GPT-6 improvements to latency- and budget-sensitive applications.

Improves substantially over Opus 5 while reducing standard token prices to $4/$20 and retaining a million-token context window.

Moves Grok beyond 4.6 with a larger base model and sizeable gains on long-running coding and professional-work benchmarks while keeping the same standard API price.

Adds a live interaction tier to the Gemini 3.8 family.

Brings deeper deliberation to Google’s low-latency live model line.

Raises coding and agentic performance while reducing inference cost and retiring the prior V4 Flash aliases.

Provides a sovereign translation option for evaluation and deployment on controlled infrastructure.

Advances Google’s dedicated generative-music model line.

Establishes the top GPT-6 capability tier with a million-token context window.

Updates the Flash line with stronger everyday reasoning and tool use while preserving its speed-oriented position.

Adds a specialised security tier to Google’s newest Flash family with controlled access for higher-risk capabilities.

Updates the Mythos branch to the current first-party version.

Replaces the original Fable 5 entry with the current first-party model version.

Brings the 5.3 family to higher-throughput workloads while remaining openly available.

Introduced native vision before the later V4.1 Flash production release.

Improves Z.ai’s coding benchmark by 50% over GLM-5.2 using scaled post-training.

Targets coherent five-minute structure, expressive intent, realistic instruments and natural vocals.

Moves the Flash tier forward from 3.6 to 3.7 and gives developers a newer low-latency default with an introductory price window.

Keeps xAI’s rapid frontier-model cadence visible while full model-card, benchmark and pricing details remain to be verified.

Unifies multimodal understanding, generation and editing for commercial video workflows.

Pushes a fast model into long-horizon agents and computer-use workflows without Pro-tier pricing.

Pairs deep software-engineering ability with an advertised serving speed of 80 tokens per second.

Brings stronger multi-agent work and programmatic tool coordination across ChatGPT, Codex and the API.

Marks Meta’s return to frontier APIs with native multi-agent orchestration and active context management.

Shows a compact model moving beyond screens into autonomous navigation across real environments.

Brings performance close to Opus 4.8 at a lower price, with stronger planning and long-running execution.

Combines a stable million-token window with multiple reasoning-effort levels and openly released weights.

Makes sovereign coding agents practical: 30B total parameters, but only 3B active at inference time.

Introduces a capability tier above Opus for complex, days-long and asynchronous tasks.

Combines open weights, a million-token context window and long-horizon agentic execution in one model.

Keeps compatibility with both OpenAI- and Anthropic-style interfaces, lowering the cost of switching.