April 2026: AI’s Biggest Month — Gemma 4, Llama 4 & Claude Mythos
2 min readApril 2026 is shaping up to be the most consequential month in the history of artificial intelligence. Within the span of two weeks, Google, Meta, and Anthropic have each shipped landmark models that redefine what open and proprietary AI can do — and the race is far from over.
Google Drops Gemma 4: Open-Source Models That Punch Far Above Their Weight
On April 2, Google DeepMind released Gemma 4, a family of four open models — 2B, 4B, 26B (MoE), and 31B — all under the permissive Apache 2.0 license. That licensing change alone is a big deal: unlike earlier Gemma versions, these models can be used commercially without restriction.
The performance numbers are striking. The 31B dense variant currently ranks #3 among all open models on the Arena AI leaderboard, while the 26B MoE variant holds the #6 spot — both outcompeting models more than 20 times their parameter count. Gemma 4 ships with context windows up to 256K tokens, native vision and audio processing, and fluency in over 140 languages. For developers building local, offline, or resource-constrained applications, this is the most capable option that has ever existed at this size.
Meta’s Llama 4 Redefines Open-Weight Multimodal AI
Three days later, on April 5, Meta launched Llama 4 Scout and Llama 4 Maverick — the first open-weight models built on a native Mixture of Experts (MoE) architecture with full multimodal training baked in from day one.
Scout is the headline grabber: its 10-million-token context window is the longest of any publicly available model, opening up workflows like reasoning over entire codebases, parsing months of user activity, or summarizing entire document libraries in a single call. Maverick, with 400 billion total parameters and 17 billion active, targets enterprise-grade reasoning and has already been benchmarked above GPT-4o on coding and long-context tasks.
Anthropic’s Claude Mythos Arrives — For a Select Few
On April 7, Anthropic quietly previewed Claude Mythos to roughly 50 partner organizations through Project Glasswing. Described internally as a “step change” above Claude Opus 4.6 — itself the top-ranked model on many benchmarks since February 2026 — Mythos is initially focused on cybersecurity vulnerability detection, advanced reasoning, and code generation. Broader access has not yet been announced.
Why This Month Matters
What makes April 2026 historic is not just the volume of releases, but the convergence of open and proprietary capability. Two of the three landmark launches this month — Gemma 4 and Llama 4 — are fully open and commercially permissive. Developers now have access to models that would have been considered state-of-the-art proprietary systems just a year ago, available for free download and local deployment.
For businesses, researchers, and builders, the calculus on AI adoption just changed. The question is no longer whether open-source models are good enough — it’s which one fits your use case.