Open models hold just 10% of global AI inference spending, and bets are on that changing fast
sarahookr · x · 2026-09-21
Konstantin Pilz highlighted a narrative-violating stat: open models still account for only 10% of global AI inference spending. Sarah Hooker quote-reposted it saying she can't wait to see the plot in 12 months, implying she expects the share to shift dramatically. The data shows that despite open-source momentum, commercial inference spend remains heavily concentrated in closed models.
More from Models
- SWE-2 free across Devin Cloud Agents, CLI and Desktop until October 8 — silasalberti · 2026-09-21
- Hebrew prompts jailbreak AI safety filters? Hands-on test says no — karminski3 · 2026-09-21
- $5, 10-minute SFT on Qwen3.6-35B-A3B lifts GPQA +8% and MMLU-Pro +12% — josh_wills · 2026-09-21
- Users fight Claude's return to mannered prose and overly safe outputs — ColleenMBrady · 2026-09-21
- How could a model spontaneously write jailbreak language? Reddit probes OpenAI's injection report — sivadneb · 2026-09-21
- GLM 5.3 Flash, DeepSeek V4.1 Flash and Qwen 3.8 all beat 'frontier' models from just 10 months ago — burny_tech · 2026-09-21