😺DeepSeek’s new 28-cent agent model
The Neuron · Grant Harvey · 2026-08-02
DeepSeek upgrades its V4-Flash model into a stronger coding and agent platform at $0.28 per million output tokens, pressuring frontier AI labs like OpenAI and Anthropic on price while posting competitive benchmark scores.
Extraction
Topics: deepseekai-model-pricingcoding-agentsmulti-agent-workflowsai-infrastructure-spending
Claims
- DeepSeek re-trained V4-Flash without increasing model size, activating approximately 13B of 284B parameters per request to keep inference costs low.
- V4-Flash scored 82.7 on Terminal-Bench 2.1 and 54.4 on DeepSWE, with Artificial Analysis raising its Intelligence Index score 10 points to 50.
- Pricing remained at $0.14 per million input tokens and $0.28 per million output tokens, making routine agent workloads dramatically cheaper than comparable frontier models.
- Big Tech has collectively spent over $1.1T on AI infrastructure since 2023, with another $745B projected for 2026.
- The 'Gauntlet Loop' multi-agent prompt technique — using specialist builder agents evaluated by separate critic agents against real-world reference examples — produced a community of 27 playable browser games from a single three-paragraph prompt using Claude Opus 5.
Key quotes
A workflow that felt too expensive at scale via OpenAI or Anthropic can suddenly make sense.
The next model war will not be won by one leaderboard. It will be won when buyers ask why a routine task still needs the expensive option.
Capability got cheaper. The systems, permissions, and power bills around it did not.