Your AI agent would cost 260% more without prompt caching, Arena data shows
arena · x · 2026-10-30
Agent Arena analyzed 26,225 sessions (30 agents, 13 labs, Sep 12–Oct 23, 2026) plus six labs' cache pricing since January 2024, explaining why OpenAI's new 12-hour API caching and never-expiring Codex/Work caches matter so much.
- Cache discount went from zero to 95% in two years: cache reads now cost 5% of input price at leading labs, vs. parity in early 2024.
- Caching keeps agents affordable: without it, a typical top-10 agent task costs 3.6x more (+260%).
- Charts show 86–96% of task tokens are re-reads; GPT-6 Astra's typical task at $4.71 would reach $801 in the priciest no-cache scenario; Claude Fable 5.1 $7.44 → $783; Kimi K3.1 Pro $1.09 → $106.
More from Infra
- effect-harness on Cloudflare: run agent loops, tools and credentials on separate VMs — samgoodwin89 · 2026-10-31
- Google's Project Suncatcher boots TPU in orbit and beams back first AI response — sundarpichai · 2026-10-31
- Together AI launches Post-Training platform for agentic RL with tool-use environments — togethercompute · 2026-10-31
- General Compute raises $100M for inference capacity as CoreWeave stock hits 12-month low — YvesMulkers · 2026-10-31
- Analyst warns of unstable AI compute feedback loop as Kioxia's NAND price warning triggers flash crash — bookwormengr · 2026-10-31
- SOMA launches enterprise prompt-compression layer claiming 20%+ inference savings — markjeffrey · 2026-10-31