Top AI Product

DeepSeek is back, and it’s aiming straight at the agent crowd. The new V4 series is two open-weight MoE models — V4-Pro (1.6T total, 49B active) and V4-Flash (284B total, 13B active) — both shipping with a 1-million-token context window by default. Not a premium tier. The floor.

What it actually is

These are API-callable frontier models built for long-horizon agentic work — coding loops, multi-step tool use, the kind of task that eats context and never lets go. V4-Pro-Max hits 80.6% on SWE-bench Verified, the top open-weights score, tied with Gemini 3.1 Pro. V4-Flash matches Pro on simpler agent tasks at 12× less cost, so most people will run Flash and only escalate when they need the muscle.

Why it matters

Pricing is the weapon. Flash lists at $0.14/M in, $0.28/M out; Pro’s cache-hit rate drops to $0.003625/M — brutal for agent loops that resend the same context every turn. DeepSeek also rolled out peak-hour surge pricing and reportedly raised $7.4B at a $50B+ valuation. The playbook is the same as V3: commoditize the model layer, force everyone to compete on cost. Long-context agents just got a lot cheaper.

You Might Also Like

• Tempo Machine Payments Protocol 500m 100 Partners and a Bold Play to own the ai Agent Economy
• 397 Billion Parameters on a 48gb Macbook Flash moe Turns Apples 2023 Research Into Reality
• Microsoft Agent Governance Toolkit Scores 10 10 on Owasp Agentic Risks at 0 1ms per Check
• Deepseek v4 pro Hits gpt 5 Parity on 5 of 7 Benchmarks at a Fraction of the Cost
• Deepclaude Lets Claude Code run on Deepseek v4 pro 0 87 vs 15 per Million Tokens