Is M2.7 better because it is newer?
Not on the two things recorded here. M2.7 costs the same and runs a 200K window against 1M for M2.5.
MiniMax M2.5 is available on A2Agent as minimax-m2.5 at $0.300 per 1M input tokens, with a 1M context window, through one OpenAI-compatible endpoint. Coding work at a 1M window on the flat MiniMax rate.
| Model id | minimax-m2.5 |
|---|---|
| Platform | MiniMax |
| Context window | 1M |
| Tagged for | chat, coding |
| Discount | … |
| Input / 1M tokens | $0.300 |
| Output / 1M tokens | $1.20 |
Prices synced with platform billing (USD per 1M tokens).
curl https://a2agent.me/v1/chat/completions \
-H "content-type: application/json" \
-H "authorization: Bearer sk-your-a2agent-key" \
-d '{
"model": "minimax-m2.5",
"messages": [{"role": "user", "content": "Reply with OK"}]
}'A cheap default for Cline and Kilo Code when the repository is large. Because all three MiniMax models cost the same, switching to M3 for an agent run costs nothing but the model string.
All three MiniMax models sold here list at the same price: $0.30 per million input tokens and $1.20 output. Price cannot decide between them, so the split is window and tag. M2.5 gives you 1M and the coding tag. M2.7 is the later release but drops to a 200K window, also coding-tagged. M3 keeps 1M and carries the agent tag instead. That makes M2.5 the pick for code that has to see a lot at once, because the newer M2.7 costs exactly the same and gives you a fifth of the context. With the rate flat across the family, moving between them is free. What you trade is what the model was tuned for and how much it can read.
Not on the two things recorded here. M2.7 costs the same and runs a 200K window against 1M for M2.5.
1M tokens.
Same window and same price. M2.5 carries the coding tag, M3 the agent tag.
At list, yes: $0.30 per million input and $1.20 output for M2.5, M2.7 and M3 alike.
Trial credit on sign-up, billed per token, no minimum spend.