AI Models News

Terminal-Bench 4.0 chart on a dark star-edged background: GPT-6 Astra 57.9%, Claude Fable 5.1 55.8%, GPT-5.6 Sol 37.3%, as reported by OpenAI.
AI Models Launch

OpenAI Releases GPT-6 Astra

OpenAI released GPT-6 Astra on September 3, 2026, a computer-use flagship at $10/$50 per million tokens with 1.05M context, Critical cyber designation under its Preparedness Framework, and staged rollout to Plus, Pro, Business, Enterprise, API, Azure, and Bedrock.

By Matthew Lake
Bar chart of Anthropic-reported Terminal-Bench 4.0 scores: Claude Mythos 5.1 at 60.9%, Claude Fable 5.1 at 55.8%, Claude Fable 5 at 42.0%, GPT-5.6 Sol at 37.3%.
AI Models Launch

Anthropic Releases Claude Fable 5.1 and Claude Mythos 5.1

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026, at unchanged $10/$50 pricing with a 75% cheaper cache read and a government-vetted Mythos access program after Fable 5's June export-control suspension.

By Matthew Lake
GPT-5.6 Family Debuts, Then OpenAI Cuts Luna's Price 80%
AI Models Launch

GPT-5.6 Family Debuts, Then OpenAI Cuts Luna's Price 80%

OpenAI shipped the GPT-5.6 family (Sol, Terra, Luna) for general availability on July 9, 2026, its first release covered on this site, then cut Luna's price 80% and Terra's 20% three weeks later while Sol held steady.

By Bobby Smart
Claude Sonnet 5's $2/$10 Pricing Ends August 31, Rises 50%
AI Models Launch

Claude Sonnet 5's $2/$10 Pricing Ends August 31, Rises 50%

Anthropic released Claude Sonnet 5 on June 30, 2026, calling it the most agentic Sonnet model yet. Introductory pricing of $2/$10 per million input/output tokens holds only through August 31, 2026, then rises to $3/$15.

By Bobby Smart
Claude Opus 5: Same $5/$25 Price, Cache-Safe Tool Swaps for Agents
AI Models Launch

Claude Opus 5: Same $5/$25 Price, Cache-Safe Tool Swaps for Agents

Anthropic released Claude Opus 5 on July 24, 2026, holding the $5/$25-per-million-token price it has charged since Opus 4.6, with beta features for mid-conversation tool changes that preserve the prompt cache and automatic fallbacks for safety-flagged requests.

By Bobby Smart
Bright pixel-art isometric scene of a powerful robotic minotaur breaking free from containment, glowing neon amber and cyan against a dark navy background, bold amber kicker reads FABLE 5 UNLEASHED
AI Models Launch

Claude Fable 5: Anthropic's First Public Mythos-Class Model Unleashed

Anthropic launched Claude Fable 5 on June 9, 2026, the first generally available Mythos-class model, priced at $10/$50 per million tokens with safety classifiers that reroute sensitive requests to Claude Opus 4.8.

By Bobby Smart
Pixel-art isometric scene: a large blue-and-white friendly robot launches upward from a sky platform on twin rocket exhausts, while three smaller matching robots look on from below; towers and white clouds fill the bright blue sky background. Kicker OPUS 4.8 LANDS in white at the bottom.
AI Models

Claude Opus 4.8: Anthropic's New Flagship Lands with Parallel Subagents

Anthropic released Claude Opus 4.8 on May 28, 2026: same $5/$25 pricing, a research-preview dynamic workflows tool for parallel subagents in Claude Code, a 3x-cheaper fast mode, and measurable alignment gains.

By Bobby Smart
Gemini 3.1 Flash Lite: Google's Cheapest AI Model Yet
AI Models

Gemini 3.1 Flash Lite: Google's Cheapest AI Model Yet

Google released Gemini 3.1 Flash Lite on March 3, targeting lightweight and edge use cases. Here's what it's designed for and where it fits in the model lineup.

By Caliph Herald
Qwen3.7-Max: Alibaba's Agent-First 1M-Context Flagship
AI Models

Qwen3.7-Max: Alibaba's Agent-First 1M-Context Flagship

Alibaba announced Qwen3.7-Max in May 2026: an agent-first flagship with a 1M-token context window and native extended thinking. Here's what the benchmarks show and how the pricing works.

By Bobby Smart
Mistral Small 4: 119B Parameters, 6B Active, Europe's Efficient AI Play
AI Models

Mistral Small 4: 119B Parameters, 6B Active, Europe's Efficient AI Play

Mistral released Mistral Small 4 on March 16, a 119B/6B active parameter MoE model combining instruction following, reasoning, vision, and coding. European data residency is a key differentiator.

By Bobby Smart
DeepSeek V4: A 1.6T Open-Weight MoE with 1M Context
AI Models

DeepSeek V4: A 1.6T Open-Weight MoE with 1M Context

DeepSeek released V4 in April 2026: V4-Pro and V4-Flash, both open-weight under MIT with a 1M-token context window. Here's what the benchmarks show and how the economics work.

By Bobby Smart
Gemini 3.1 Pro: Google Takes the SWE-Bench Crown
AI Models

Gemini 3.1 Pro: Google Takes the SWE-Bench Crown

Google released Gemini 3.1 Pro on February 19 with an 80.6% SWE-bench score, 94.3% GPQA Diamond, and 77.1% ARC-AGI-2. Here's what those numbers mean and how it compares.

By Bobby Smart
A pixel-art llama wearing a small hat, carrying a brown travel suitcase, raising one hoof in a farewell wave as it walks away down a winding dirt road into bright sunny rolling green hills at golden hour
AI Models

Meta's Muse Spark: The End of Open-Weight Llama

Meta Superintelligence Labs shipped Muse Spark on April 8, 2026: a proprietary multimodal reasoning model that marks Meta's exit from open-weight Llama. Here's what it does and why the strategy shift matters.

By Bobby Smart
Claude Sonnet 4.6 Drops: Faster Output, Same Opus Intelligence
AI Models

Claude Sonnet 4.6 Drops: Faster Output, Same Opus Intelligence

Anthropic released Claude Sonnet 4.6 on February 17, two weeks after Opus 4.6. It runs faster and costs less while matching Opus on most tasks. Here's what's different.

By Bobby Smart
Claude Opus 4.6: Anthropic's 1M Context Window Changes Everything
AI Models

Claude Opus 4.6: Anthropic's 1M Context Window Changes Everything

Anthropic released Claude Opus 4.6 with a 1 million token context window, 128K output tokens, and an 80.8% SWE-bench score. Here's what those numbers mean for developers.

By Bobby Smart