Models

Earlier MiniMax generation, still very cheap per token.

M2.5 HighSpeed for tight latency budgets.

Balanced M2.7 for general chat and code assist.

M2.7 tuned for latency. Same cost as the quality path.

Top MiniMax line. Cheap long-context agent work.

Lightest Gemini. Classification, extraction, bulk transforms.

Gemini 3 generation Flash. Long context, flash pricing.

3.7 Flash for tools that pin that revision. Same economics.

A million tokens of context for the price of a rounding error.

Prior Grok revision for pinned toolchains.

Strong reasoning, unusually willing to argue with your premise.

The dependable middle. Rarely the best answer, rarely a bad one.

The default for real work. Long context, steady tool use.

Previous-gen workhorse. Still solid, slightly cheaper than 5.6.

Deliberate reasoning mode. Slower, and worth it on hard bugs.

Preview Pro. Strong multimodal reasoning before it hardens.

Best-in-class on long agentic runs. Priced like it knows that.

Frontier OpenAI. The one you reach for when nothing else lands.

Fast Claude. The one you leave running in a loop.