This guide uses source checks from Sep 13, 2026. Provider and gateway prices can change; preserve the checked date with every forecast.
Why This Topic Matters Now
The last week of AIWave SEO posts leaned into DeepSeek, Qwen, GLM, Kimi, and brand search cleanup. MiniMax deserves a separate treatment because it is not just another row in a comparison table. The current public MiniMax pages describe multimodal foundation models with code and agent capabilities, ultra-long context handling, and model coverage that includes M3 and M2-family routes. That makes MiniMax a route-ledger problem for teams evaluating agent workloads.
A MiniMax trial should answer whether the route is a good fit for a specific agent class: coding review, document synthesis, multimodal reasoning, support drafting, or long-context planning. A live price row alone cannot answer that. The ledger needs source dates, pricing_version, route identity, cache behavior, output caps, retry boundaries, and acceptance evidence. That is the difference between testing a model and approving a production route.
Source Facts Checked Today
AIWave /api/pricing checked on Sep 13, 2026 returned success=true, 64 model rows, pricing_version a42d372ccf0b5dd13ecf71203521f9d2, auto_groups=['default'], group_ratio default=3 and vip=1, and public rows enabled for default, vip, and svip groups. Computed base examples per 1M text-token units were DeepSeek Flash at $0.70 input, $0.02331 cache-hit input, and $2.10 output; DeepSeek V4 Flash at $0.638 input, $0.020288 cache-hit input, and $1.914 output; MiniMax M2 at $0.46866 input, $0.046866 cache-hit input, and $1.87464 output; MiniMax M3 at $0.90608 input, $0.181216 cache-hit input, and $3.62432 output; Doubao Seed 2.0 Mini at $0.178537 input and $1.785369 output; Doubao Seed 2.1 Pro at $1.339027 input and $6.695133 output; Qwen3.8 Max at $2.678053 input and $8.034160 output; and Qwen Image 3.0 Pro at $0.055793 input/output unit. Actual invoices still depend on account group, final route, request usage, and the receipt captured for that run.
AIWave /api/pricing checked on Sep 13, 2026 exposed MiniMax M2 at $0.46866 input, $0.046866 cache-hit input, and $1.87464 output per 1M units; MiniMax M2.7 at $0.45304 input, $0.090608 cache-hit input, and $1.81216 output; and MiniMax M3 at $0.90608 input, $0.181216 cache-hit input, and $3.62432 output. Highspeed variants appear as separate rows and should not be merged with standard routes.
Public MiniMax pages checked through search results describe M3 and M2-family coverage, multimodal capabilities, code and agent use cases, and long-context positioning. Those pages are useful capability context, while AIWave's live pricing endpoint remains the gateway evidence for the rows used in this article.
Planning Matrix
A source-dated planning matrix keeps the page useful for engineers and procurement reviewers. It turns a search query into an auditable route decision instead of a loose model preference.
| MiniMax route | Useful first trial | Ledger field to inspect |
|---|---|---|
| MiniMax-M2 | General agent drafting | cache_hit_input and output cap |
| MiniMax-M2.7 | Engineering planning | reviewer score and retry count |
| MiniMax-M2.7-highspeed | Latency-sensitive triage | route label and timing context |
| MiniMax-M3 | Multimodal or coding-agent review | context size and acceptance notes |
| MiniMax highspeed rows | Interactive workflows | do not blend with standard rows |
| Fallback route | Incident continuity | error class and allowed action |
Implementation Pattern
The implementation pattern keeps credentials as placeholders, pins the AIWave base URL, records the model, and leaves room for route-specific controls. Production applications should move credentials into environment or secret storage.
from openai import OpenAI
client = OpenAI(api_key="YOUR_API_KEY_HERE", base_url="https://aiwave.live/v1")
ROUTE_PLAN = {
"agent_draft": {"model": "MiniMax-M2", "max_tokens": 700},
"agent_review": {"model": "MiniMax-M3", "max_tokens": 1000},
}
def run_minimax_route(task_name: str, prompt: str):
route = ROUTE_PLAN[task_name]
response = client.chat.completions.create(
model=route["model"],
messages=[{"role": "user", "content": prompt}],
max_tokens=route["max_tokens"],
temperature=0.15,
)
return {
"model": route["model"],
"api_key": "YOUR_API_KEY_HERE",
"pricing_checked_at": "2026-09-13",
"pricing_version": "a42d372ccf0b5dd13ecf71203521f9d2",
"usage": response.usage,
}
Begin With Agent Classes
MiniMax route selection should begin with agent classes rather than model popularity. Define a small set of real tasks: coding plan, code review, support draft, document synthesis, multimodal inspection, and incident triage. Each class needs an expected answer shape, maximum output tokens, retry budget, and reviewer. The same prompt set should run across candidate routes so quality and usage are comparable. Without that structure, a stronger-looking answer can hide a route mismatch.
Do Not Blend Standard and Highspeed Rows
Highspeed route names should be treated as separate products in ledgers. A highspeed row can be reasonable for interactive workflows, but it should not be averaged with a standard row in a monthly forecast. Store the exact model string, route speed class, source date, and pricing_version. If the product switches from MiniMax M2.7 to MiniMax M2.7-highspeed for a user-facing feature, that is a route change requiring review.
Measure Cache Fields by Template Version
MiniMax rows in the live AIWave table expose cache-related ratios for several routes. That is useful only if the workload repeats enough stable context. Store prompt_template_version, stable_prefix_hash, context size, and cache-hit input when available. If an agent dynamically rewrites its instructions or includes changing file maps, the cache forecast should be marked unproven until repeated runs show stable behavior.
Control Long Context
Long-context models tempt teams to send every file, every ticket, and every policy excerpt into each request. A ledger should separate required context, optional context, and retrieved context. Use a context budget per task class. A coding review may need a module map and the changed files, while an architecture review may need a broader reference pack. The route should pass acceptance within the assigned budget before production promotion.
Map Errors to Actions
Agent workloads need explicit failure policy. Authentication errors, insufficient balance, rate limits, upstream timeouts, malformed payloads, and user cancellations all need different actions. A MiniMax fallback to another MiniMax row may be acceptable for drafting, while coding review might require a hard stop and human review. Store error_class, fallback_allowed, fallback_model, retry_count, and final_status in the ledger.
Internal Links for MiniMax Evaluators
This article should link to Models docs, Chat Completions, Pricing, Pricing JSON, Status, and Trust. The MiniMax page also creates a fresh internal-link destination for provider-specific buyers who are not searching for DeepSeek, Qwen, GLM, or Kimi.
Procurement Review
Procurement should ask for model string, speed class, checked date, pricing_version, account group, context budget, output cap, cache measurement, retry budget, and acceptance owner. If a team cannot answer whether a row is standard or highspeed, it should not yet approve that route for production traffic. The finance review should preserve live AIWave row evidence beside the engineering acceptance notes.
Final Checklist
A MiniMax route ledger is ready when agent classes are defined, standard and highspeed rows are separated, cache behavior is measured, context budgets are enforced, errors map to actions, and the reviewer has accepted quality for each task class. Recheck live AIWave pricing before rollout and store a dated copy of the route receipt with the release notes.