SambaNova serves its model list without an API key, and there are only 7

Almost every inference provider puts /v1/models behind auth. Groq returns 403 without a key, Together returns 401, DeepSeek returns 401, Cerebras returns 403.

SambaNova just answers:

curl -s https://api.sambanova.ai/v1/models

No key, no account, no header. As of 2026-08-28 it returns exactly 7 models:

DeepSeek-V3.1                  131072
DeepSeek-V3.2                   32768
Meta-Llama-3.3-70B-Instruct    131072
MiniMax-M2.7                   196608
MiniMax-M3                    1048576
gemma-4-31B-it                 131072
gpt-oss-120b                   131072

Seven is a small number, and that is the useful part. Provider comparison posts tend to quote catalog sizes in the hundreds, which mostly measures how many old checkpoints someone left listed. This is the whole menu, and you can see it before you sign up for anything.

Two details in there that surprised me:

MiniMax-M3 carries a 1,048,576-token context window. That is a million tokens on a provider whose whole pitch is throughput rather than context.

DeepSeek-V3.2 is listed at 32,768 — a quarter of V3.1's 131,072. Newer version, smaller window. If you upgrade the model string on the assumption that context goes up or stays flat, that one will bite.

Since the endpoint needs no credentials, it also makes a decent CI check: fetch it, diff the id list against whatever your config hardcodes, fail the build when a model you depend on stops being listed. Most providers make you burn a key to do that.

More on what SambaNova actually gives you for free: SambaNova Cloud free API