promptgarten 🌱

📊 Model Timeline

When did which model arrive — and when does something disappear? The chronicle of major AI models: releases, sunsets, and price changes, each with a source.

ReleaseSunset / DeadlinePrice change
2026-08-31Anthropic

Claude Sonnet 5 — end of introductory pricing

The introductory price of $2/$10 per million tokens ends; after that the regular price of $3/$15 applies.

Claude Docs: Pricing

2026-07-24DeepSeek

DeepSeek: old model names get migrated

From this day, the old API names deepseek-chat and deepseek-reasoner point to V4-Flash — anyone using the old models should check their integration.

DeepSeek API Docs: Pricing

2026-07-24Anthropic

Claude Opus 5

Anthropic released Claude Opus 5: the same price as Opus 4.8 ($5/$25 per million tokens), a 1M-token context, thinking enabled by default, a five-level effort control, and a fast mode at double the price ($10/$50). Since Claude Code v2.1.219 (July 24, 2026), Opus 5 is the new standard Opus there; Opus 4.8 remains in use as an automatic fallback for safety-related refusals.

Anthropic: Claude Opus 5

2026-07-20Anthropic

Claude Fable 5 — permanent in Max & Team Premium

From 20 July 2026 included in all Max and Team Premium plans (at 50% of limits). Pro and Team Standard keep access via usage credits and receive a one-time $100 credit. The previously announced sunset did not happen.

Anthropic announcement (@claudeai, via Simon Willison)

2026-07-16Moonshot AI

Kimi K3

2.8 trillion-parameter MoE with 1M context — according to Moonshot the largest open model in the world; the weights follow on July 27, until then API-only.

Moonshot: Kimi K3

2026-07-16xAI

Grok 4.5

Positioned for agentic coding, 500K context, $2/$6 per million tokens — the cheapest model in xAI's own comparison chart.

xAI: Grok 4.5

2026-07-15Thinking Machines Lab

Inkling (Thinking Machines)

First open-weights model from Mira Murati's lab: an MoE with 975B total / 41B active parameters, 1M context, 77.6% SWE-bench Verified — billed as the strongest open US model. A smaller preview variant, Inkling-Small, has 12B active parameters.

Thinking Machines Lab: Introducing Inkling

2026-07-09Meta

Muse Spark 1.1 + Meta Model API

Meta's first closed-weight frontier model, accessible via the new OpenAI-SDK-compatible Meta Model API (public preview).

Meta AI Blog: Muse Spark

2026-07-09OpenAI

GPT-5.6 family (Sol, Terra, Luna)

Three tiers: Sol as the flagship ($5/$30), Terra as the mid-range option ($2.50/$15), and Luna as the cheapest variant ($1/$6).

TechCrunch: GPT-5.6-Launch

2026-06-18Google

Gemini CLI + Code Assist for GitHub discontinued

Google is discontinuing Gemini CLI for free users; the successor is the Antigravity CLI. Code Assist for GitHub: deprecation starts June 18, full shutdown July 17.

9to5Google: Gemini CLI Shutdown

2026-05-20Alibaba

Qwen3.7-Max

Alibaba's agentic flagship ("designed for the agent era") — proprietary for the first time instead of open-weight, with 1M context.

Alibaba Cloud Blog: Qwen3.7

2026-05-19Google

Gemini 3.5 Flash (GA)

According to Google, 4x faster output and ahead of its own 3.1 Pro on coding benchmarks — Flash is no longer a budget tier ($1.50/$9).

Google Blog: Gemini 3.5

2025-12-22Zhipu / Z.ai

GLM-4.7

358B MoE under the MIT license — an open-weights workhorse for agentic coding, with a free Flash variant.

Z.ai Blog: GLM-4.7

2025-12-02Mistral

Mistral Large 3

Europe's open-weights flagship: 675B MoE under Apache 2.0, with vision and an EU hosting option.

Mistral News: Mistral 3