Clef Flash
The 9B low-latency Clef — sub-40ms decisions at $0.038 per million tokens.
About
Clef-flash, released October 1, 2026 alongside Clef, is Cloudflare's latency-tier decision model: 9 billion parameters on a Qwen3.5-9B backbone, built for the hot path of agent pipelines — routing support tickets, filtering bot traffic, triage — where tens of milliseconds matter more than maximum accuracy.
Cloudflare reports a median latency of 38.8 ms, roughly 13x faster than TypeSafe's Jev in its own tests. On October 9, 2026, Cloudflare cut the price from $0.09 to $0.038 per million input tokens — undercutting Jev's $0.042 — with output still free. The same update limited the hosted context window to 24K tokens (down from 64K); the open weights still support 256K tokens for self-hosting under Apache 2.0.
It speaks the same Jev-compatible System One API as Clef, so the two can be mixed: flash for volume, full Clef for the hard calls.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported: median 38.8 ms (~13x faster than Jev's 524.1 ms). Independent third-party bench (Oct 10, 2026): 65/82 correct, median 0.521 s over the network.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Release facts (October 1, 2026; 9B on Qwen3.5-9B; 38.8 ms median; $0.09→$0.038 price cut October 9; 24K hosted / 256K self-host context) corroborated across Cloudflare's changelog and independent coverage. Docked for: (1) latency and accuracy figures are Cloudflare's own; (2) the hosted context cut to 24K limits long-state decisions unless you self-host; (3) the price war it started means today's rate card may not hold.