* nvidia: add kimi-k3 and deepseek-v4-flash-0731 provider entries
Both models are live on the NVIDIA NIM hosted catalog
(integrate.api.nvidia.com/v1) but were missing from the nvidia
provider, so opencode and other consumers show an outdated picker.
Provider files follow the override-only convention:
base_model + cost + reasoning_options + interleaved.
* fix(nvidia): address review — kimi-k3 wire schema, trial-tier cost citation
- kimi-k3: match peer control set (toggle + effort low/high/max) and
document the exact NIM wire paths in the required leading comment.
Verified empirically against integrate.api.nvidia.com/v1:
chat_template_kwargs.thinking toggles reasoning on/off, top-level
reasoning_effort accepts low|high|max.
- deepseek-v4-flash-0731: keep 0.0 pricing with a comment citing the
NVIDIA API Trial Terms of Service free tier.
* fix(nvidia): single reasoning_options form; cite model card for cost
- kimi-k3: collapse the duplicate reasoning_options key (assignment +
table-array header) into one valid inline array; control set unchanged
(toggle + effort low/high/max).
- deepseek-v4-flash-0731: replace the generic trial-ToS note with a direct
citation of this ID's catalog card
(https://build.nvidia.com/deepseek-ai/deepseek-v4-flash-0731).
* docs(nvidia): lead with source comments per AGENTS.md
Move the NIM wire-schema and pricing-source notes into a leading header
above the first key in both new entries, and state the free trial-tier
posture explicitly for deepseek-v4-flash-0731 (source: its catalog card).
---------
Co-authored-by: Not-Saii <Not-Saii@users.noreply.github.com>
saii
d1b84739db
