Signal65: token prices rise but cost per finished AI task stays flat as agent accuracy jumps
ryanshrout · x · 2026-09-12
Signal65's analysis on its PINNACLE benchmark pushes back on the "AI is getting more expensive" narrative: priced per finished task, costs are flat.
- Newest closed models finish 98%+ of multi-step enterprise jobs with few hallucinations
- Top PINNACLE score moved 3.6x in two weeks; top-tier cost per finished task rose only 11%
- GPT-6 lists at 2.5x GPT-5.6 but costs the same per finished job; Gemini 3.8 Flash makes 11x fewer errors than the best Gemma for 1.5x the cost
Bottom line: price per token went up, price of finished work did not — agentic capability is improving faster than pricing. The benchmark uses deterministic code-verified scoring, with NVIDIA Blackwell Ultra and AMD as platform partners.
Related event: Signal65's PINNACLE Benchmark: AI Cost Per Completed Task Hasn't Risen(2 posts)→
More from Models
- Qwenputer: a full RISC-V CPU inside Qwen3.5-4B weights, running C in forward passes — armand_ruiz · 2026-10-31
- OpenAI blocks weapons details for teens, adds auto-ending chats fixated on violence — rohanpaul_ai · 2026-10-31
- Granite Switch 4.1.1 ships: update models like you update software — krvarshney · 2026-10-31
- Mistral Open-Sources Le Chonk: Top Open Model on FrontierCode, Cyberdefense-Tested by CrowdStrike and Cloudflare — tensorqt · 2026-10-31
- Stanford CS336 Lecture Explains Why DPO Compares Against a Reference, Not Two Current Probabilities — stanfordnlp · 2026-10-31
- ik_llama.cpp PR #2808 adds native runtime for China Telecom's Xing4.0-29B-A4B MoE with mHC and MLA — Any_Bus9205 · 2026-10-31