April 20, 2026 – AI Daily Recap

The AI industry enters the new week with a flurry of model releases, open-source breakthroughs, and billion-dollar power plays. Anthropic dropped its most capable public model yet, Google pushed open-weight intelligence to the edge, DeepSeek is days away from unveiling a trillion-parameter challenger built on Chinese silicon, and an independent team published an open reconstruction of Claude Mythos that punches well above its weight class. Here’s everything that matters.

Anthropic Ships Claude Opus 4.7 With High-Res Vision and Stronger Agentic Coding

Anthropic released Claude Opus 4.7 on April 16, delivering a 13% lift on coding benchmarks, three times more production tasks resolved autonomously, and the first Claude model to support high-resolution image input at 2,576 pixels — roughly triple the visual capacity of previous generations. Pricing stays flat at $5 per million input tokens and $25 per million output tokens, making it the first Opus upgrade that doesn’t raise the bill. Anthropic also introduced new cybersecurity safeguards that automatically detect and block prohibited high-risk requests, alongside a new Cyber Verification Program for legitimate security researchers. The model is live across the API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.

Google Launches Gemma 4: Frontier Open Models That Run on a Raspberry Pi

Google released Gemma 4 on April 2, a family of four open-source models — from a tiny 2B effective-parameter variant up to a 31B dense powerhouse — all released under the Apache 2.0 license. Built from the same research that produced Gemini 3, these models offer native vision and audio processing, 256K context windows, and fluency in over 140 languages. The headline achievement: a 31B-parameter model that beats 400B rivals on key benchmarks while fitting on a single 80GB GPU. The smallest variants run entirely offline on phones, Raspberry Pi, and NVIDIA Jetson boards with near-zero latency — a genuine milestone for on-device AI.

DeepSeek V4 Imminent: A Trillion-Parameter Model on Huawei Chips

DeepSeek V4 is expected to launch in the latter half of April with roughly one trillion total parameters, of which only 32–37 billion activate per task through a Mixture-of-Experts architecture. Reuters confirmed the model will run on Huawei’s Ascend 950PR chips, making it the first frontier AI model built to run entirely on Chinese semiconductor infrastructure. With a one-million-token context window and an anticipated Apache 2.0 license, V4 is positioned to challenge GPT-5.4 and Gemini 3.1 Pro at a fraction of the training cost — DeepSeek’s previous model reached competitive performance for an estimated $5.2 million.

OpenMythos: An Open-Source Reconstruction of Claude Mythos at Half the Parameters

Independent researcher Kye Gomez published OpenMythos on April 19, an open-source PyTorch reconstruction of what Anthropic’s unreleased Claude Mythos architecture might look like. The project proposes a Recurrent-Depth Transformer with Mixture-of-Experts routing, and its 770M-parameter prototype matches the performance of a 1.3B fixed-depth Transformer trained on the same data. The architecture uses Linear Time-Invariant injection to prevent residual explosion and Adaptive Computation Time halting to avoid “overthinking.” This is not a leaked model or distillation — it’s a falsifiable hypothesis rendered in code, and it’s already sparking debate about whether the gap between frontier and open models is smaller than assumed.

EU Locks In AI Hiring Rules With 105 Days to Go

The European Union’s AI Act hiring provisions are now 105 days from taking effect, and the compliance clock is ticking for employers using algorithmic screening. The rules classify AI systems used in recruitment and worker management as “high risk,” requiring mandatory bias audits, transparency obligations, and human oversight before any automated hiring decision becomes final. Companies that fail to comply face fines of up to 3% of global annual revenue. The announcement comes as the U.S. takes a sharply different approach: the White House’s National Policy Framework explicitly recommends against creating new federal AI rulemaking bodies and is pushing for federal preemption of state-level AI laws.

Q1 2026 Venture Funding Shatters Records as AI Absorbs 81% of All Capital

Global startup funding hit $297 billion in Q1 2026, with AI companies claiming $242 billion — 81% of all venture capital deployed worldwide. Four of the five largest venture rounds in history closed in a single quarter: OpenAI at $122 billion, Anthropic at $30 billion, xAI at $20 billion, and Waymo at $16 billion. Foundational AI startup funding in Q1 alone was double all of 2025. Meanwhile, OpenAI has crossed $25 billion in annualized revenue with 900 million weekly active users and is taking early steps toward a public listing, while Anthropic’s annualized revenue has reportedly surpassed $30 billion.

The Big Picture

The defining tension of April 2026 is between openness and control. Google and DeepSeek are shipping frontier-capable models under permissive licenses that run on commodity hardware, while Anthropic withholds its most powerful model (Mythos) entirely — and an open-source team immediately attempts to reverse-engineer it. The venture funding numbers are staggering but increasingly concentrated: four companies absorbed the lion’s share of $242 billion in a single quarter. And as the EU tightens AI hiring rules, the U.S. moves to preempt its own states from doing the same. The race isn’t just about who builds the best model anymore — it’s about who sets the rules for how they’re deployed.

Sources

• Anthropic — Introducing Claude Opus 4.7
• Tom’s Guide — I Tested Anthropic’s Claude Opus 4.7
• AI Business — Anthropic Releases Claude Opus 4.7
• Google Blog — Gemma 4
• Engadget — Google Releases Gemma 4
• Tech Insider — Gemma 4: How a 31B Model Beats 400B Rivals
• Hugging Face — Welcome Gemma 4
• GizChina — DeepSeek V4 Expected to Launch in Late April
• FindSkill.ai — DeepSeek V4: Release Date, Specs, and the Huawei Chip Bombshell
• SitePoint — DeepSeek V4 Released: What’s New
• GitHub — OpenMythos
• MarkTechPost — Meet OpenMythos
• The Decoder — The Myth of Claude Mythos Crumbles
• Asanify — EU Locks In AI Hiring Rules
• Consumer Finance Monitor — White House AI Policy Framework
• Crunchbase — Q1 2026 Shatters Venture Funding Records
• Crunchbase — Foundational AI Startup Funding Doubled
• OpenAI — Accelerating the Next Phase of AI