Arena data: Gemini 4 Carbon scales cheapest on hard tasks, rivals cost 3-4x more
arena · x · 2026-11-05
New Agent Arena cost-scaling data shows Google's Gemini 4 Carbon holds its cost edge as tasks get harder: on typical tasks GPT-6 Astra costs 3.4x, Claude Opus 5.5 3.9x, and GPT-6.1 Sol about 0.54x as much. In the priciest 5% of sessions Carbon's position improves across the board, and OpenAI/Anthropic drop from 10 of 11 frontier spots on cheap tasks to 7. Over 1,000 sessions crossed $100 in the last two weeks, so suboptimal model choice meaningfully inflates bills.
Related event: Arena Cost Curves Show Gemini 4 Carbon Leads on Hard Tasks(2 posts)→
More from Infra
- Proving a Model's Answer Costs 70-95x the GPU Time of Running It on vLLM — petrusenko_max · 2026-11-05
- AWS shows how to plug calibrated decision models into Bedrock AgentCore Evaluations — AWS ML Blog · 2026-11-05
- LFM3-Omni-2B ported to run fully offline on iPhone at ~80 tok/s with custom WebGPU kernels — JosephJacks_ · 2026-11-05
- Orbital Data Centers Launch to Sun-Synchronous Orbit, Memed as Elves Sailing West — rand_longevity · 2026-11-05
- Vietnam's Lâm Đồng to bring Starlink to mountain areas where poor signal hampers rescue — elonmusk · 2026-11-05
- SpaceX partners with Qualcomm for nationwide Starlink Mobile 5G network from 2027 — Polymarket · 2026-11-05