Latest Artificial Intelligence News & Technical Dispatches

Meta releases Llama 4.5 with 405 billion parameters, 410 tok/s throughput, 128K context window, and Apache 2.0 license. At $0.15 per million tokens on API providers, Llama 4.5 reshapes the economics of open-weight AI for enterprise deployments.

Speculative decoding has matured from research paper to production standard in 2026. Medusa heads, Eagle draft models, and self-speculation techniques cut inference latency by 2.5x without quality loss. Deep dive into architecture, benchmarks, and deployment trade-offs.

OpenCode's launch hit 1,274 HN points — the highest for any AI coding tool in 2026. 14,200 GitHub stars in 48 hours, 7K-token system prompt (79% less than Claude Code), and 47 community MCP integrations.

LiteLLM (1.8ms overhead, 200+ providers) vs Unify (3.2ms, dynamic latency benchmarking). Which proxy should power your production AI stack? Complete benchmark data, config files, and decision matrix.

Docker Sandboxes GA'd September 1 with 180ms cold starts, gRPC connection pool API, and native OpenCode integration. $0.002/min with volume discounts at 10K+ minutes/month.

Enterprise LLM costs dropped from $200 to $30 per million tokens through 5 optimization layers. This production playbook covers prompt compression, semantic caching, speculative decoding, multi-model routing, and batch processing with runnable TypeScript code.

Nat Friedman's Entire platform (611 HN points) aims to be the GitHub for AI agents — hosted MCP registry, agent manifest spec, and sandboxed execution runtime.

HelixDB combines vector search and graph traversal in a single Rust engine with 4.2ms hybrid queries. This deep dive covers the LSM tree architecture, HNSW index configuration, Apache Arrow memory model, and scaling benchmarks from 1K to 1M nodes.

Anthropic shipped three production agent capabilities to general availability on August 19-20, 2026: Computer Use (desktop control), Browser Use (in-page automation), and Tool Search Tool (85% context savings). This is the most significant agent infrastructure release of 2026 — the complete production agent stack in a single GA window.

Google shipped Gemini 3.7 Flash on August 13, 2026 — just three weeks after 3.6 Flash — at half the price ($0.75/1M input) with 340 tok/s throughput. The model achieves 43.6% on FrontierCode 1.1 Main and 65.3% on DeepSWE v1.1, making it the most cost-effective model for production agentic coding pipelines.

Production agents need memory beyond the context window. This analysis compares three memory patterns — short-term (Redis buffers), long-term (vector stores), and episodic (graph databases) — with production benchmarks on latency, cost, recall accuracy, and failure modes across 120K+ daily agent sessions.

The EU AI Act's first enforcement deadline hit on August 2, 2026 — banning prohibited AI practices and triggering high-risk system requirements. AI developers building systems used in the EU must implement risk classification, data governance, transparency obligations, and human oversight mechanisms before the February 2027 full enforcement date.

Anthropic's Tool Search Tool, shipped to GA on August 19, 2026, fundamentally changes how agents consume tool definitions. Instead of loading all 50+ MCP tool definitions upfront (consuming 72K+ tokens), agents discover tools on-demand, loading only the 3-5 relevant definitions. Opus 4 accuracy improved from 49% to 74% on MCP evaluations.

Google shipped Gemini 3.7 Flash on August 13, 2026 at half the price of 3.6 Flash — $0.75/1M input tokens with 340 tok/s throughput. This deep dive benchmarks its coding, reasoning, and agentic capabilities against Claude 3.7 Sonnet, GPT-5.6, and open-weight alternatives to determine where Flash wins and where it falls short.

CrowdStrike launches Falcon IQ with 50+ Charlotte AI agents and NVIDIA Nemotron models at Fal.Con 2026, expanding Project QuiltWorks across 12+ partners.