Clef-flash
Released Oct 1, 2026 · knowledge cutoff unpublished
- Status
- Unknown
- Location
- United States
- Type
- System One
- Modality
- Multimodal
- Context window
- 66K
- Max output
- —
- Speed
- —
- Price ($/MTok in / out)
- $0.09 / —
- Cost per task
- —
- Open weights
- Yes
Strengths & weaknesses
Strengths
- Answers in a few tens of milliseconds at the median, fast enough to sit inline on a request path rather than in a background queue
- Same Jev-compatible API and typed-question schema as Clef, so the two can be swapped per call to trade accuracy for latency
- Takes images and video as state, not just text and JSON
Weaknesses
- Shares the user-reported hosted-API state plateau of about 2,048 tokens seen on Clef, well short of the advertised window
- Calibration is Cloudflare's own claim with no independent audit yet
Not an LLM: the 9B sibling of Clef, post-trained from Qwen3.5-9B with the same schema head, returning probabilities for typed noul, choice and score questions in one forward pass, so none of this site's benchmarks apply and every cell is null by design (Cloudflare reports a decision-format GPQA Diamond of 51.0, not comparable). Workers AI lists $0.09 per M input tokens (8,182 neurons) and no output rate; outputPerMTok is null for the reason given on Clef. Cloudflare reports a 38.8 ms median latency against Jev's 524.1 ms, its own figure. Released with Clef on 2026-10-01. No Artificial Analysis coverage; knowledge cutoff unpublished.
Benchmarks0 of 10 reported
MMLU-Proretired
GPQA Diamond
SWE-bench Verified
SWE-bench Pro
Terminal-Bench 2.1
AIME (latest)retired
Humanity's Last Exam
LMArena Elo
GDPval-AA v2
ARC-AGI-2
For developers
API model strings
- cloudflare
@cf/cloudflare/clef-flash
Licence
Apache License 2.0 · Permissive · commercial use permitted
Retirement
No retirement announced
Lineage
No recorded predecessor