Open models hit 78.4% of Vercel AI Gateway tokens; DeepSeek and Moonshot spend top OpenAI
realmeetjames · x · 2026-09-21
- rauchg's data: On Vercel's AI Gateway, open-weight models took 78.4% of daily token volume vs 21.6% closed. By spend, Moonshot AI and DeepSeek rank #3 and #4, and combined with Z.ai their inference spend exceeds OpenAI (#2) — though that's inference spend across providers (mostly US), not revenue to the open labs.
- sovereignbrah's take (quoted): Anthropic and OpenAI's doomsday warnings reflect a lost market capture — enterprises don't need pricey near-AGI models when open weights deliver 95% of capability at 1-2% of the price. He argues they underestimated competition, their IPOs are cooked, and they're now pushing a regulatory cartel to ban rivals. Opinion, but paired with the Vercel data it marks a notable shift toward open models in real workloads.
Related event: Open-Weight Models Hit Record 78.4% of Tokens on Vercel AI Gateway(10 posts)→
More from Models
- Chinese AI labs spotted scraping the web through Singapore IPs — gaganghotra_ · 2026-09-21
- ChatGPT can pinpoint any city from a photo, wilderness within 50-100 miles — menhguin · 2026-09-21
- X Users Claim "Opus 5.5" Output Beats Rivals, But No Official Release Exists — DavidmComfort · 2026-09-21
- Developer reports Astra feels "lobotomized" since Saturday, burning weekly quota on off-spec tasks — almmaasoglu · 2026-09-21
- Jev beats 7 frontier models on speed and cost ($0.004 vs $1.11), yet all score coin-flip on user preference — PrimaryMagician · 2026-09-21
- $0.63 CLINC150 eval of Jev: excellent calibration (ECE 0.025) but the fallback bucket can't be trusted — Artistic-Blood7501 · 2026-09-21