Progress in complex, long-running agentic tasks is better measured by tokens consumed rather than raw time. Improving token efficiency, as seen from GPT-5 to 5.1, directly enables more tool calls and actions within a feasible operational budget, unlocking greater capabilities.

The new multi-agent architecture in Opus 4.6, while powerful, dramatically increases token consumption. Each agent runs its own process, multiplying token usage for a single prompt. This is a savvy business strategy, as the model's most advanced feature is also its most lucrative for Anthropic.

The focus in AI engineering is shifting from making a single agent faster (latency) to running many agents in parallel (throughput). This "wider pipe" approach gets more total work done but will stress-test existing infrastructure like CI/CD, which wasn't built for this volume.

Jensen Huang reframes AI compute as a productivity investment, not a cost. He would be "deeply alarmed" if a $500,000 engineer used less than $250,000 in tokens, comparing it to a chip designer refusing to use CAD tools. This sets a radical new benchmark for leveraging AI in high-skilled roles.

Ramp's CPO argues companies shouldn't excessively worry about AI token costs. If an AI agent can deliver 10x the output of a human, it's logical and profitable to pay the agent (via tokens) more than the human's salary. This reframes ROI from a cost center to a massive productivity investment.

In the AI era, token consumption is the new R&D burn rate. Like Uber spending on subsidies, startups should aggressively spend on powerful models to accelerate development, viewing it as a competitive advantage rather than a cost to be minimized.

The binary distinction between "reasoning" and "non-reasoning" models is becoming obsolete. The more critical metric is now "token efficiency"—a model's ability to use more tokens only when a task's difficulty requires it. This dynamic token usage is a key differentiator for cost and performance.

Obsessing over linear model benchmarks is becoming obsolete, akin to comparing dial-up speeds. The real value and locus of competition is moving to the "agentic layer." Future performance will be measured by the ability to orchestrate tools, memory, and sub-agents to create complex outcomes, not just generate high-quality token responses.

In complex, multi-step tasks, overall cost is determined by tokens per turn and the total number of turns. A more intelligent, expensive model can be cheaper overall if it solves a problem in two turns, while a cheaper model might take ten turns, accumulating higher total costs. Future benchmarks must measure this turn efficiency.