▌system-one

~ / compare

compare

The 97 models in the list, side by side. A dash means the list does not say, not that the answer is no. Open an entry for the full description and its sources.

97 of 97
modelaccess$/M inlatencyrankapitakessizecontextlicense
agent-jev–––––––––
Ainvest-Decision-27b–––––––––
basal-omni-1.0-ultraopen–––yestxt img12B–31B–Apache-2.0
Chon-1.0-12Bopen–~0.11 s p50 (p95 23.5 s)–yestxt12B6K state / 8K promptApache-2.0
Clefopenhosted$0.038~130 ms––txt img aud vid9B–30B64KApache-2.0
CLMopen–––yes–8B–Apache-2.0
Cua-S1open–––––706K––
d1openhosted–~8 ms#5–txt img aud600M–3B––
d3–––––––––
deciderhosted–––yes–27B––
Decisioopen–~1.1 s for 50 questions––txt12B–31B––
dohnutsopen––––txt img0.8B up–Apache-2.0
Drexopenhosted$0.05––yestxt img vid9B–35B131KOpen RAIL-M
FlexDM-LoRA 2.1–––––––––
FRIDA-Decisionsopen––––txt–––
GEV-26B-Decideopen–~45 ms (System 1)–yestxt img26B-A4B (4B active)–Apache-2.0
GLiNER2.5-Decideopen––––txt340M–1B–Apache-2.0
GUI-Decisionsopen–~150 ms per step––txt img2B–31B––
H2O-Lightning-4Bopen––#7yestxt img4B–31B–Apache-2.0
Intern-Decisionopen––––txt img0.8B–4B–Apache-2.0
Jebadiah v2.1open–125 ms median (9B)––txt9B–27B–Apache-2.0
Jeffan-31b-1710–––––––––
Jev-Omniopen––––txt img aud vid12B–Apache-2.0
Jevaopen––––txt4B–Apache-2.0
JevEmbedopen––––txt––Apache-2.0
JevK5open–~13 ms (H100)–yestxt4B–9B–Apache-2.0
jevlikeopen––––txt–––
jevosopen–––yestxt~1B–MIT
jiwoopen–~50 ms (H100)–yestxt0.8B–4B–Apache-2.0
julia-1open–CPU-class––txt144M––
kevopenhosted–––yestxt0.8B–27B–Apache-2.0
Kev-Bio–––––––––
KnowLine 31B-Gen1open–~255 ms p50#83–txt31B–Apache-2.0
Layaopen–~33 ms––txt––Apache-2.0
liquamen–––––––––
Lumma-Fev-27Bopen––––txt27B–Apache-2.0
MiniCPM-OCR-Jev–––––––––
mammoth-1 / elephant-1openhosted––#9–txt9B–27B–MIT
NanoJevopen––––txt0.6B–MIT
NeuDecidegated––––aud~40 MB––
Nimbleopenhosted$0.04–#8yestxt9B––
OneJevopen––––txt img vid0.8B–27B–Apache-2.0
Open Jevopen––––txt150M––
openJev-verdict-2.0open––––txt151M––
Phocinaeopen––––txt144M––
Piccolo-27B–––––––––
PlayJevopen––––img0.8B–Apache-2.0
Quyetopen––––txtTiny–31B–Apache-2.0
REAX 1.3–––––––––
ReJevopen––––txt2B–MIT
RSI-Jevopen–––yestxt img27B–MIT code, Apache-2.0 weights
Sarvam Prism–––––––––
Sieve–––––––––
Smaug-Decisions–––––––––
StartLux-Decisionopen––––txt0.8B–27B–Apache-2.0
Strands Decideropen–––––E2B–26B–Apache-2.0
Surogate Runeopenhosted––––txt img26B-A4B–Gated (HF terms)
system-one-open–––––––––
Tacit / Tacitusopen––––txt1.7B–31B––
Tev1openhosted––––txt0.8B–4B–MIT
THX-01–––––––––
Torchcast Decisionopen–33 ms per noul (H100)–yestxt img4B–27B262KCC BY-NC 4.0
Turion Jev 31B–––––––––
Valenopen––––txt img vid–––
vegamlopen––––txt img800M–4B73K–
Velaopen–––yestxt0.3B–9B–Apache-2.0
VideoJev–––––––––
WaterSheepopen–––yestxt––Apache-2.0
WeMM-JEV–––––––––
xoropen–––yestxt img vid35B-A3B–Apache-2.0
Celeris-1closedhosted$0.04–#16–txt img–––
Cicero–––––––––
Decider–––––––––
Decider 1closedhosted$0.04––yestxt–4K–
Gemini Decision 1–––––––––
GLiDEclosedhosted––#4–txt–––
Hanzo Kaihosted–––yes––––
Instinctclosedhosted––#25––4B–27B––
Jevclosedhosted$0.04270–500 ms#15yes––32K–
Jev Routerhosted––––––––
ketteiclosedhosted$0.06~0.2 s#3yestxt img–––
Mercury Decidehosted––#14yes––––
Microsoft-Decision-1hosted––#19––9B––
Myles-Gclosedhosted––#2–txt12B–31B––
OpenAI Decisions APIhosted–150–230 ms#24–txt img–––
Parallel Decisionsclosedhosted$1.00200 ms–1 s––txt–32K per question–
Reflex–––––––––
Sageclosedhosted––#11–––––
Snowflake Decisionclosedhosted–seconds per call (throughput-shaped)––txt–––
Solar Decidehosted–––yestxt–512K–
Span-01hosted––––––––
SPX-CDhosted––#23yes––––
SubQ Snaphosted$0.05–#10–––384K–
Super Decideropenhosted––#1–txt26B-A4B (NVFP4)–Apache-2.0
Trio-Sparkclosedhosted$0.042––yestxt img vid–––
wity-1closedhosted––#20–––––
Clairicclosedhosted$0.10–––txt img–64K–

how to read this

  • takes: txt text, img image, aud audio, vid video.
  • api: the list says the model speaks TypeSafe's/v1/systemone request shape, so a client built for Jev works by repointing the base URL.
  • rank: position on JevBench's hosted board (of 235), as reported in the entry. Benchmarks differ between vendors; open the entry before leaning on a number.
  • $/M in: US dollars per million input tokens, as quoted in the entry.
  • Figures are read from each entry's description as of 2026-10-31. Spot something stale? Fix the listand this table follows.