Add SambaNova Cloud (20 RPM / 20 RPD / 200K TPD, no card)

[xyzs996]

Adds SambaNova Cloud to Inference providers. data.json only — the README diff above is what CI will regenerate.

Why it qualifies

contributing.md rule	this entry
Permanent free tier, no trial credits, no time-limited promos	The Free Tier is defined by the absence of a linked payment method, not by a clock. Nothing expires.
No credit card required to sign up	Correct — adding a card is what moves you off the Free Tier, into the Developer Tier.
REST API for text inference, not just a playground	OpenAI SDK-compatible at https://api.sambanova.ai/v1.

Official docs confirming the free tier and its limits

SambaNova model rate limits — the Free Tier
table, quoted verbatim: "Applied when there is no payment method linked with your account."
All five free models carry the same allowance:

Model	RPM	RPD	TPD	status
DeepSeek-V3.1	20	20	200,000	production
Meta-Llama-3.3-70B-Instruct	20	20	200,000	production
gpt-oss-120b	20	20	200,000	production
DeepSeek-V3.2	20	20	200,000	preview
gemma-4-31B-it	20	20	200,000	preview

I listed the three production models in the table and put the two preview ones in the footnote,
since the docs mark those "for evaluation and experimentation, not production".

Context and Max Output are not in the rate-limits page, so I read them from SambaNova's public,
unauthenticated catalogue on 2026-08-24 — anyone can re-check this without an account:

$ curl -s https://api.sambanova.ai/v1/models | jq -r '.data[] | [.id, .context_length, .max_completion_tokens] | @tsv'
DeepSeek-V3.1                   131072  7168
DeepSeek-V3.2                    32768  7168
Meta-Llama-3.3-70B-Instruct     131072  3072
gemma-4-31B-it                  262144  262144
gpt-oss-120b                    131072  131072
MiniMax-M2.7                    ...
MiniMax-M3                      ...

MiniMax-M2.7 and MiniMax-M3 are served but appear only in the Developer-tier tables, so they
are not in this entry.

Two things worth flagging rather than hiding

• gpt-oss-120b Max Output is —. The catalogue reports max_completion_tokens equal to
  context_length for it, which reads as "no separate output cap" rather than a real 131K ceiling.
  Rather than print a number I could not source, I left the cell blank, the way the SiliconFlow row
  does. Same situation for gemma-4-31B-it, which is footnote-only anyway.
• RPM and RPD are both 20. That is unusual enough to look like a typo, but it is what the table
  says: one busy minute spends the entire day's request budget. I put that sentence in the footnote
  because it changes how you'd actually use this tier.

What I could not verify

I have not made an authenticated call against the free tier — no account, so no live probe like the
one behind the Kilo Code footnote. Everything above is from SambaNova's published docs and their
public catalogue endpoint, both linked. If you want a live confirmation before merging, that part
is not something I can supply.