August 31, 2026 — curated links and takeaways.
1. DeepSeek V4 Flash Leads Cost Competition at $0.14/$0.28 Per Million Tokens
DeepSeek V4 Flash is now the cheapest production-grade API at $0.14 input / $0.28 output per million tokens, while DeepSeek V4 Pro remains permanently 75% discounted. This pricing floor forces all vendors downward and shifts cost economics for high-volume inference workloads.
2. Cashfree Relay: Payment Automation Agent Exits Beta, Goes GA
Cashfree Payments launched Relay, an AI-powered Super Agent for payment operations automation, moving from merchant beta (May 2026) to general availability. This signals enterprise adoption of task-specific agents in fintech workflows where automation ROI is measurable.
3. OpenAI Astra Resolved Ten Open Math and Theory Problems Pre-Launch
OpenAI announced on August 1 that Astra—its unreleased frontier model—solved ten long-standing problems in mathematics and theoretical computer science. This capability claim raises the bar for reasoning-class models and signals OpenAI's competitive positioning ahead of release.
4. Gemini 3.5 Pro Missed Three Release Deadlines; Google Engineering Prioritizes Search Over Agents
Gemini 3.5 Pro missed release targets in June, mid-July, and August with internal sources attributing delays to Google's divided focus between search integration and agent capabilities. This signals execution risk at Google and an opening for competitors in the agent-prioritized market.
5. Llama 4 Maverick Now Cheapest Flagship-Class Model at $0.15 Per Million Input Tokens
Meta's Llama 4 Maverick is priced at $0.15 per million input tokens—the lowest among production flagship-class models, below GPT-5.6 and Claude tier pricing. This cost advantage for developers choosing mid-tier models reshapes build decisions for latency-tolerant workloads.