Cloudflare, OpenAI Launch Decision Model Alternatives to TypeSafe Jev
Twenty new decision models have shipped in four weeks to compete with TypeSafe's Jev, offering open weights and multimodal support.
Key points
- TypeSafe launched Jev on September 15, 2026, as a closed-source decision model.
- At least 20 alternatives from major vendors and developers shipped within four weeks.
- Cloudflare’s Clef outperforms Jev on specific benchmarks but lags on others.
TypeSafe introduced Jev on September 15, 2026, establishing a new category of decision models that return structured labels or probabilities rather than generating free-form text. Within approximately four weeks of this launch, at least twenty competing alternatives entered the market, according to MarkTechPost. These new models come from major technology companies and independent developers, signaling rapid competition in the structured AI output space.
The competitors include offerings from Cloudflare, OpenAI, Perplexity, Liquid AI, and AWS’s Strands Labs. MarkTechPost reports that these twenty models aim to address specific limitations found in Jev’s initial release. Developers have cited gaps such as the lack of image support, the absence of self-hosting options, undisclosed model sizes, and the inability to fine-tune Jev. The new alternatives attempt to fill some or all of these functional voids for engineering teams.
Benchmark Performance and Model Specifications
Performance comparisons vary significantly depending on the benchmark suite used, with most figures being self-reported by the vendors. MarkTechPost notes that scores across different suites like JevBench and Respan are not directly comparable. However, Cloudflare’s Clef model scored 94.20 on the BANKING77 F1 metric in its own run, surpassing Jev’s reported score of 79.74. Conversely, Jev maintained a lead on the GPQA Diamond benchmark, scoring 78.3 compared to Clef’s 48.0.
Other models show wider performance disparities. The Von 1.2 encoder scored 0.422 accuracy in Respan’s evaluation, while Jev achieved 0.932 on the same test. Model sizes for the alternatives range from 287 million parameters for GLiNER2.5-multi-Decide to 35 billion parameters for Mapika Decider. TypeSafe has not disclosed the size of Jev. Most open models trail Jev on complex reasoning tasks but match or exceed it on narrow classification problems.
Vendor Offerings and Licensing
Cloudflare announced Clef on October 1, 2026, providing two variants: the 27-billion-parameter Clef and the 9-billion-parameter Clef-flash. Both models are available under the Apache-2.0 license on Hugging Face and are served via Workers AI. They support text, JSON, and up to four images per request, addressing Jev’s text-only limitation. Pricing for Workers AI is $0.24 per million input tokens for Clef and $0.038 for Clef-flash.
Liquid AI released its d1 family, which includes hosted options and two open-weight models. The d1-3B model runs on LFM2.5-VL-3B, while the d1-omni-600M variant adds support for audio clips up to thirty seconds. These open weights are distributed under the LFM Open License. Some models, like Von and Laya, can run on CPUs, while larger adapters require GPUs with 24 GB of memory or equivalent Mac hardware.
What should you do?
Security and technology professionals should evaluate whether their current workflows require multimodal inputs or self-hosting capabilities that Jev currently lacks. Teams should verify benchmark results against their specific use cases, as performance varies widely between reasoning and classification tasks. It is important to review licensing terms for any open-weight models considered for production use.
What to do and how to stay safe: TypeSafe
- Assess whether your application requires image or audio processing, which Jev does not support but some alternatives do.
- Compare pricing structures, noting that Cloudflare’s Clef-flash is cheaper per token than Jev’s listed rate.
- Verify hardware requirements for self-hosted models, as some require specific GPU memory configurations.
- Review license agreements for open-weight models to ensure compliance with your organization’s software policies.
Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Does Jev support image inputs?
No, TypeSafe’s documentation states that Jev accepts text only, which is one of the gaps developers have raised.
Are there free output tokens for Jev?
Yes, TypeSafe’s docs list $0.042 per 1M input tokens with free output tokens included in that pricing structure.
Which model has the best BANKING77 F1 score?
Cloudflare’s Clef scored 94.20 on the BANKING77 F1 metric in Cloudflare’s own run, compared to Jev’s 79.74.
