This year
The most recent entry is GLM-5.3 from Z.ai on August 14, 2026. New launches are added to this table on release day, newest first.
| Date | Model | Provider | Why it mattered |
|---|---|---|---|
| August 14, 2026 | GLM-5.3 | Z.ai | The GLM line's August refresh in a crowded open-weights month. |
| August 13, 2026 | Gemini 3.7 Flash | The fast tier keeps absorbing capability between Pro releases. | |
| August 6, 2026 | Grok 4.6 | xAI | Focused on long-running agents, with a 500K context window. |
| August 2, 2026 | Qwen3.8 Max | Qwen | Alibaba's biggest Qwen release of the summer. |
| July 16, 2026 | Grok 4.5 | xAI | xAI's mid-year coding and cost push. |
| July 16, 2026 | Kimi K3 | Moonshot | A 2.8T open-weights flagship announced with weights following on July 27. |
| July 9, 2026 | GPT-5.6 (Sol, Terra, Luna) | OpenAI | Three named variants replace the single-flagship pattern. |
| June 30, 2026 | Claude Sonnet 5 | Anthropic | The 5-family workhorse: near-flagship quality at Sonnet pricing, default across Anthropic products. |
| June 9, 2026 | Claude Fable 5 | Anthropic | The first Mythos-class model, opening Anthropic's 5 family above the Opus tier. |
| May 28, 2026 | Claude Opus 4.8 | Anthropic | The Opus line's spring capstone. |
| April 24, 2026 | DeepSeek V4 | DeepSeek | DeepSeek's next major line lands one day after GPT-5.5. |
| April 23, 2026 | GPT-5.5 | OpenAI | Billed as OpenAI's smartest and most intuitive model; Instant variant reached free users in May. |
| April 16, 2026 | Claude Opus 4.7 | Anthropic | First of three frontier releases inside eight days. |
| March 5, 2026 | GPT-5.4 | OpenAI | The 5-series cadence settles at roughly one release per quarter. |
| February 19, 2026 | Gemini 3.1 Pro | Google's spring flagship refresh. | |
| February 5, 2026 | GPT-5.3-Codex | OpenAI | A dedicated coding flagship, aimed at agentic software work. |
2025: the reasoning race and the price collapse
The year DeepSeek R1 made cheap reasoning a market event, reasoning models became the default frontier format, and November packed four frontier releases into thirteen days.
| Date | Model | Provider | Why it mattered |
|---|---|---|---|
| December 11, 2025 | GPT-5.2 | OpenAI | OpenAI's December refresh of the 5-series. |
| December 1, 2025 | DeepSeek V3.2 | DeepSeek | The sparse-attention release goes stable, with tool calls in reasoning mode. |
| November 24, 2025 | Claude Opus 4.5 | Anthropic | Anthropic closes the November wave with a large price cut on its flagship tier. |
| November 18, 2025 | Gemini 3 Pro | Google's November entry in the densest release month the field had seen. | |
| November 12, 2025 | GPT-5.1 | OpenAI | Instant and Thinking variants, opening a packed November. |
| September 29, 2025 | DeepSeek V3.2-Exp | DeepSeek | Sparse attention in the open: the experiment that became V3.2. |
| August 7, 2025 | GPT-5 | OpenAI | The long-awaited flagship, with automatic routing between fast and thinking modes. |
| July 11, 2025 | Kimi K2 | Moonshot | A trillion-parameter open-weights model from China's Moonshot, strong on agentic tasks. |
| July 9, 2025 | Grok 4 | xAI | xAI reaches the frontier tier, with live X data as the differentiator. |
| May 22, 2025 | Claude Opus 4 and Sonnet 4 | Anthropic | Positioned squarely at coding and long-running agent work. |
| April 16, 2025 | o3 and o4-mini | OpenAI | Reasoning models get tool use: thinking plus searching, coding, and reading images. |
| April 5, 2025 | Llama 4 (Scout, Maverick) | Meta | Meta's open-weights line goes multimodal with very long context. |
| March 25, 2025 | Gemini 2.5 Pro | Google takes benchmark leads with a 1M context window as the headline feature. | |
| February 24, 2025 | Claude 3.7 Sonnet | Anthropic | Hybrid reasoning: one model deciding when to think longer. |
| January 20, 2025 | DeepSeek R1 | DeepSeek | The release that moved markets: open-weights reasoning at roughly a tenth of Western API prices. |
The build-up: 2022 to 2024
The releases that created the current landscape: the ChatGPT launch, the first frontier race, and the arrival of multimodal and reasoning models.
| Date | Model | Provider | Why it mattered |
|---|---|---|---|
| December 26, 2024 | DeepSeek V3 | DeepSeek | Open weights near frontier quality at a fraction of the training cost. The stage-setter for January. |
| December 11, 2024 | Gemini 2.0 Flash | Fast-tier models get genuinely capable, foreshadowing the cheap-model price collapse. | |
| September 12, 2024 | o1-preview | OpenAI | The first mainstream reasoning model: slower answers that think before responding. |
| June 20, 2024 | Claude 3.5 Sonnet | Anthropic | A mid-tier model beating the previous flagship, resetting expectations for the Sonnet tier. |
| May 13, 2024 | GPT-4o | OpenAI | Native voice and vision in one model, and free-tier access to frontier capability. |
| March 4, 2024 | Claude 3 (Opus, Sonnet, Haiku) | Anthropic | The tier naming that stuck: a flagship, a workhorse, and a fast cheap model. |
| Date | Model | Provider | Why it mattered |
|---|---|---|---|
| December 6, 2023 | Gemini | Google's answer to GPT-4 and the start of the three-way frontier race. | |
| July 11, 2023 | Claude 2 | Anthropic | Anthropic opens Claude to the public, with a then-huge 100K context window. |
| March 14, 2023 | GPT-4 | OpenAI | The first frontier model most professionals used, and the start of paid AI subscriptions at scale. |
| Date | Model | Provider | Why it mattered |
|---|---|---|---|
| November 30, 2022 | ChatGPT | OpenAI | The launch that made conversational AI a consumer product. |
What the cadence tells you
The gaps keep shrinking. GPT-4 held the frontier for over a year. By late 2025, four labs shipped frontier models in a single month, and in April 2026 three landed inside eight days. A subscription decision made on "which model is best" now has a shelf life measured in weeks.
Releases cluster on purpose. Labs increasingly ship into each other's launch windows: Claude Opus 4.7, GPT-5.5, and DeepSeek V4 arrived within eight days of each other in April 2026, and the November 2025 wave was even denser. Launch-week benchmark charts are marketing artillery in those windows, which is why they rarely settle anything.
The open-weights track is no longer behind. DeepSeek, Kimi, Qwen, GLM, and Llama entries have moved from footnotes to frontier contention in two years, usually at a fraction of the price. The practical effect is the widening cost spread the AI Model Cost Index tracks.
The practical conclusion has not changed since GPT-4: when a release lands, do not switch on the announcement. Run your own three real tasks against the new model next to your current one; the release evaluation guide is a one-hour method for exactly that. In Whizi the new frontier models land in the same picker as the old ones, so the comparison is a conversation rather than a second subscription.
Method, sources, and corrections
Dates are public release-announcement dates, verified against provider announcements and same-day press coverage before an entry is added. Where only a month is solidly sourced, the entry shows the month without inventing a day. Preview, waitlist, and staged rollouts are dated to the first public availability announcement.
The table deliberately tracks major releases rather than every checkpoint: a release earns an entry when it changed what a mainstream user or developer could buy, run, or compare. Minor point updates and silent model swaps do not qualify.
Corrections are welcome and applied fast: if a date here disagrees with a primary source, use the contact on the support page and cite the source. This page is intended to be citable, so its accuracy is the product.
- Check the date on this page's newest entry before citing it as current
- Treat launch-week benchmark charts as marketing until third parties reproduce them
- Test a new release on three of your own tasks before switching anything
- Expect the frontier to change hands within weeks, and pick tools that let you follow it
- Cite the entry date, not the coverage date; they differ for staged rollouts
Frequently asked questions
What is the most recent major AI model release?
The newest entry in this timeline is GLM-5.3 from Z.ai, released August 14, 2026. The table above is updated on launch days, so the top row is the current answer.
How often do new AI models come out?
Fast and accelerating: the trailing twelve months of this timeline contain 22 major releases across the frontier labs and the open-weights ecosystem, and that counts only launches significant enough to earn an entry. The practical takeaway is to stop optimizing for a permanent winner.
When was ChatGPT released?
November 30, 2022. It was the moment large language models became a consumer product, and every entry in this timeline is downstream of the subscription market it created.
Should I switch models every time something new comes out?
No. Switch when a new model wins on your own repeated tasks, not on launch benchmarks. The one-hour evaluation in the release guide linked above is enough to tell, and running it inside a multi-model workspace means the test costs a conversation instead of a new subscription.