NeuralStackIn a landscape where new large language models seem to appear every few days, AI Release Tracker has quietly become a de facto scoreboard for the modern AI race. The site maintains a continuously updated timeline of every major AI model release from leading labs, starting with the launch of ChatGPT on November 30, 2022. From that single inflection point, the service now covers 199 tracked frontier models from OpenAI, Anthropic, Google, Meta, SpaceXAI, DeepSeek, Mistral, Moonshot AI, Cursor, and Z.ai.
The tracker’s homepage presents the arms race in stark visual form: a dense timeline stretching from late 2022 through 2026, overlaid with expected next releases for each of the biggest players. OpenAI, Anthropic, Google, Meta, SpaceXAI, DeepSeek, Mistral, Moonshot AI, Cursor, and Z.ai each have upcoming model dates listed, with some already overdue and others slated for release in the coming weeks and months. A handful of labs, including OpenAI, Mistral, and Cursor, show expected drops that have slipped past their target dates, underscoring how closely the community is watching launch cadence.
Underneath that timeline, the site drills down into the individual model families that have defined this era of AI. OpenAI’s section spans the evolution from GPT‑3.5 and GPT‑4 through GPT‑4 Turbo, GPT‑4o and its mini variants, and on to an extensive GPT‑5 series that includes Pro, Codex, and nano lines, capped by newer iterations such as GPT‑5.6 Sol, Terra, and Luna. Anthropic’s list traces a parallel arc for Claude, from early versions through the Claude 3 era, into the 3.5 and 3.7 upgrades, and up to the current wave of Claude 4 and 5‑class models, with Sonnet, Haiku, Opus and Fable variants marking distinct performance and capability tiers.
Google’s footprint on the tracker shows its shift from Bard to the Gemini era, cataloging Nano, Pro, Ultra, and Flash tiers across multiple generations. The chronology runs through Gemini 1.x and 2.x families, Gemma open‑weight releases, and culminates in the Gemini 3.x and 3.5 series, including Flash, Flash‑Lite, and a Cyber‑branded Flash variant. Meta’s LLaMA lineage is charted from LLaMA 1 and 2 through multiple Code Llama versions, into LLaMA 3 and 3.x iterations, and on to the LLaMA 4 Scout and Maverick models, alongside Muse Spark entries that hint at broader multimodal ambitions.
Newer entrants are given equal treatment. SpaceXAI’s Grok line is tracked from Grok‑1 through Grok‑4, Grok‑2.5, specialized coding and fast variants, and experimental beta releases like Grok 4.20 and 4.3. DeepSeek’s portfolio covers coder, math, MoE, and V‑series models, including V2, V2.5, V3 and R1‑based chat systems, up through DeepSeek‑V4‑Pro and DeepSeek‑V4‑Flash, with the latest Flash variant dated with a late‑July 2026 release. Mistral’s section is similarly dense, listing 7B and Mixtral families, Medium, Large and Small tiers, specialized Codestral and Mathstral models, and a growing set of NeMo, Pixtral and Ministral releases, culminating in third‑generation and fourth‑generation lines.
Beyond raw chronology, AI Release Tracker leans into analytics. Each individual model page includes release date, parameter count, context window, model type and benchmark scores where available, covering GPQA Diamond for graduate‑level science reasoning, SWE‑Bench Verified for real‑world software engineering, and MMMU for multimodal understanding. An Analytics page aggregates this into cumulative release charts, monthly release frequency, year‑over‑year totals, benchmark frontier progression, and company‑by‑company release cadence. The project’s data has already been cited by Business Insider, which highlighted that the monthly cadence of major AI releases has roughly quadrupled since 2023.
Why this matters
The result is that AI Release Tracker is more than just a list of model names—it is a live instrument for measuring how quickly frontier AI capabilities are changing. With the most recent tracked frontier model, DeepSeek‑V4‑Flash‑0731, landing on July 31, 2026, the timeline makes clear that top labs are pushing out new iterations at an unprecedented pace, and that “frontier” is now a moving target rather than a static milestone. By tying each release to benchmarks like GPQA Diamond and SWE‑Bench Verified—where current leaders include an OpenAI GPT‑5.4‑Pro model on science reasoning and an Anthropic Claude Opus 4.7 on software engineering—the tracker also surfaces how performance leadership can shift as fast as the releases themselves.
For practitioners, investors, and policymakers, the service’s company‑level cadence charts provide a way to see which labs are accelerating and which are stabilizing. OpenAI’s long GPT‑4 and GPT‑5 sequences, Anthropic’s dense Claude 4.x and 5 lines, Google’s Gemini progression, Meta’s LLaMA expansions, SpaceXAI’s Grok experiments, DeepSeek’s rapid V‑series iterations, and Mistral’s growing family of compact and large models collectively tell a story of escalating competition. At the same time, the tracker’s FAQ emphasizes its role as a free, neutral resource rather than a ranking: it follows releases from the first ChatGPT launch through “today,” updating the homepage and analytics as new models land.
Looking ahead, AI Release Tracker’s “expected next release” column may be one of its most revealing features. By plotting upcoming drops for OpenAI, Anthropic, Google, Meta, SpaceXAI, DeepSeek, Mistral, Moonshot AI, Cursor, and Z.ai—and flagging when those dates slip—it turns opaque roadmaps into visible signals of momentum and delay. As the number of tracked frontier models pushes beyond 199 and the monthly cadence continues to rise, the site is likely to become an even more important reference point for anyone trying to understand not just what AI can do today, but how quickly that picture is changing.