Skip to content

3.39.0 — gateway drift re-sync

Choose a tag to compare

View all tags
@VickyXAI VickyXAI released this 13 Aug 04:07
· 84 commits to main since this release
v3.39.0
f8a23d1

Franklin's free default was a dead model, and the whole catalog had drifted. The gateway (Base + Solana) shipped ~100 commits since Franklin's last sync (2026-07-25). Everything below was verified against the live gateway, not the repos' claims.

Free tier rebuilt on live probes

  • nvidia/qwen3-next-80b-a3b-instruct (the free default across ~20 files) hit NVIDIA's EOL 2026-07-27 (410) — every free turn was silently riding a pooled gateway fallback
  • Live probes: mistral-nemotron is DEGRADED (streaming calls 400), the 30B omni answers as gpt-oss-120b, step-3.7-flash leaks thinking prose
  • New default: nvidia/nemotron-nano-9b-v2 — the one free model that serves itself on the streaming path, verified end-to-end through the binary
  • Free vision turns now route to nemotron-nano-12b-v2-vl instead of vision-unsupported

Catalog & pricing

  • 7 price fixes: GPT-5.6 luna $1/$6 → $0.20/$1.20, terra cut, deepseek-chat/reasoner → $0.14/$0.28, gemini-3.5-flash 3× correction, glm-5 raise
  • 11 new chat models (GPT-5.6 Pro tiers, GPT-5.5 Pro, Gemini 3.6 Flash, Qwen3.7 Plus/Flash, Tencent HY3, Xiaomi MiMo v2.5 Pro, …)
  • Media: Seedance 2.5 + 2.0-mini, grok-imagine 720p tier, 4 new image models; dall-e-3 retired
  • Cost estimates now include the flat $0.001 gateway transaction fee

Prediction markets

  • crossPlatform → zero-cost tombstone (Predexon sunset market matching 2026-07-20; endpoint de-registered). All 9 remaining /v1/pm/* paths verified live.

Dependencies

  • @blockrun/llm 3.8.3 → 3.13.1 (Solana key-format fix, fail-fast PM helpers, Router Core V3)

646/646 local tests pass; free default verified live ($0, no fallback banner).

Assets 2
Loading