CLM-8B: A 9x Faster System 1 Decision Model Than Jev
A Medium post argues that three weeks after System One models emerged, the category already includes open-source rivals to Jev. The author compares CLM-8B, Clef and Laya on speed, vision capability, and local runnability.
A Medium post argues that three weeks after System One models emerged, the category already includes open-source rivals to Jev. The author compares CLM-8B, Clef and Laya on speed, vision capability, and local runnability.
An independent pre-registered evaluation, ASSAY-001, tested Jev's claim that it cannot hallucinate values outside a schema. Across 8,576 responses on Banking77 and CLINC150 it emitted zero schema violations, but confidence calibration failed on Banking77 (ECE 0.0936) while passing on CLINC150 (ECE 0.0204). Two additional independent evals found Jev losing to Claude Haiku 4.5 on phishing classification and to bge-m3 on reranking unless fused.
An independent pre-registered evaluation, ASSAY-001, tested Jev's claim that it cannot hallucinate values outside a schema. Across 8,576 responses on Banking77 and CLINC150 it emitted zero schema violations, but confidence calibration failed on Banking77 (ECE 0.0936) while passing on CLINC150 (ECE 0.0204). Two additional independent evals found Jev losing to Claude Haiku 4.5 on phishing classification and to bge-m3 on reranking unless fused.
A Fireship video covers Jev, a 'System 1' AI model built in stealth over two years by ex-OpenAI researcher Diogo Almeida. The model cannot talk or write code but is claimed to run 200x faster and 400x cheaper than LLMs while being hallucination-free. Its launch challenges the assumption that language is necessary for AI reasoning.
A Fireship video covers Jev, a 'System 1' AI model built in stealth over two years by ex-OpenAI researcher Diogo Almeida. The model cannot talk or write code but is claimed to run 200x faster and 400x cheaper than LLMs while being hallucination-free. Its launch challenges the assumption that language is necessary for AI reasoning.
Show HN: Gutsy, subsecond decision model inference on CPU, no API
Show HN: Gutsy, subsecond decision model inference on CPU, no API
Jeeves: reasoning improves Jev-like decision models
Jeeves: reasoning improves Jev-like decision models