SubQ Snap launches: 384K-token decision model at $0.05/M, beats Jev on Decision Index
Med1_Ai · x · 2026-10-13
SubQ Snap debuts as the first long-context "decision model" for enterprise work, with per-token compute staying constant as context grows. Key specs: 384K multimodal tokens (12x larger than Jev), 61.7 on Decision Index v0.2.1 vs Jev's 57.9 (leading on 13 of 16 enterprise benchmarks), sub-second decisions on 100K-token inputs, and $0.05/M input tokens with free output. The first 10,000 sign-ups get 100M free input tokens. The author argues RAG pipeline overengineering is unnecessary when models handle 384K multimodal context natively.
Related event: Subquadratic Launches SubQ Snap, a 384K-Context Decision Model(9 posts)→
More from Models
- Muse senior engineer benchmark results pulled over flawed run, retest underway — danshipper · 2026-10-13
- "Zero chance" a Meta model competes with Google, OpenAI or Anthropic right now — iruletheworldmo · 2026-10-13
- Dev turns EmbeddingGemma 2 into an on-device STT model, transcribing 1 hour of audio in 13 seconds — solyarisoftware · 2026-10-13
- Alexandr Wang hails Theo's model rankings: Opus 5.5 and Muse Wonder alone in S tier — alexandr_wang · 2026-10-13
- Reddit user rails against Gemini's decline: wrong bulb model, 13 failed script fixes — hotboxuzi · 2026-10-13
- steipete: 6.1 Sol ultrafast beats 6 Astra — same cost, but 250 tpm changes everything — steipete · 2026-10-13