Independent Jev reports|Not affiliated with TypeSafe AI
jevmodel.ai
Last updated:

Jev model reports for GPT, Claude, Gemini, Grok, Kimi, DeepSeek, Muse Wonder and images

Eight reports bring together X posts, follow-ups, and project documentation. Seven cover language-model families. The image report covers OCR, filenames, pixel-color questions, a separate drawing model, and Jev Day native image input behind TypeSafe-Danger-Zone-Images. The Muse Wonder report covers skills that install Jev inside Muse and Muse Code. The cited Muse workflows date from 19 September to 1 October 2026 and name Muse or Muse Code. The GPT report includes Artificial Analysis Decision Index v2 next to GPT-6 Luna (Max), ClickHouse’s 0.951 nDCG@10 warehouse table ranking, Sanity’s 0.5-second docs agent with gpt-oss-120b, and Arize’s bake-off against OpenAI Decisions. The Claude report includes RunSybil’s security triage against Opus 5.5, Sardine’s 27-of-29 fraud-ring scoring against Claude Sonnet 5, Decision Index v2 next to Haiku 5.5 (Max), and Takuma Yoshida’s 100 fictional return emails against Haiku 5.5. The Gemini report includes Salesgear’s 203-thread triage against Gemini 3.8 Flash, TypeSafe’s Jack & Jill screen against Gemini 3.1 Flash Lite, and Google’s Argon announcement, which names no Jev score. The write-ups separate runtime handoffs, side-by-side tests, and cases where a language model only helped write the code. The Jevtember page covers Vercel’s September AI Gateway adoption figures. The Jev Day 2026 hub covers jev-1.14.0, Box, and 262k context. The Composio Firewall page covers session hooks that evaluate tool calls with Jev.

OpenAI66 sources

Jev + GPT / Codex

Artificial Analysis Decision Index v2 scores Jev 1.14 at 67 and GPT-6 Luna (Max) at 76. ClickHouse ranks warehouse tables with Jev at 0.951 nDCG@10. RunSybil’s filter keeps GPT-6 Astra xhigh at 23.1% TPR against tuned Jev at 63.2%.

Read report
Anthropic50 sources

Jev + Claude

RunSybil’s tuned jev-1.13.0 filter reaches 63.2% TPR against 47.9% for Opus 5.5 xhigh on security findings. Sardine’s fraud-ring test catches 27 of 29 with Jev against 18 for Sonnet 5. Decision Index v2 scores Jev 1.14 at 67 and Haiku 5.5 (Max) at 77.

Read report
Google40 sources

Jev + Gemini

Salesgear’s 203-thread test scores Gemini 3.8 Flash at 199 of 203 and Jev 1.13.0 at 193. TypeSafe’s Jack & Jill case study reports replacing Gemini 3.1 Flash Lite on an initial candidate screen.

Read report
xAI4 sources

Jev + Grok

A 1,000-post selection workflow and a separate action-router setup, with source links and no inferred speed or cost claims.

Read report
Moonshot AI9 sources

Jev + Kimi

A measured email handoff to Kimi K3, a SemEval chart against Kimi-K2 Thinking, one snapshot where a Kimi K3 alias returned 139 valid answers of 231, and a hook log that named kimi-k3 once.

Read report
DeepSeek24 sources

Jev + DeepSeek

Vals prices DeepSeek V4.1 Flash at $0.55 per 1,000 claim-verification cases with probabilities, against $0.02 for Jev 1.13.0. Arena still compares its Jev Router with DeepSeek V4.1 Flash (Max).

Read report
Meta14 sources

Jev + Muse Wonder

Meta released Muse Wonder on 12 October 2026. The cited Muse workflows install Jev as a typed judge inside Muse or Muse Code. Those posts and READMEs date from 19 September to 1 October 2026 and name Muse or Muse Code.

Read report
Images26 sources

Jev + Images

Hosted Jev takes text by default. Image files need TypeSafe-Danger-Zone-Images on jev-1.13.0 and jev-1.14.0. The reports below also cover OCR first, filenames, pixel-color questions, and a drawing step gated by Jev.

Read report

Compare measurements within the same task and dataset. The benchmark reports preserve those boundaries; the source notes record what each author actually reported.