Minimax + LiteLLM Cost Estimation (Team)
Cost estimates for running Minimax models via a LiteLLM proxy for a development team of 20–30 people at medium daily usage.
Why / When to Use
Use these figures when planning LLM infrastructure budget for a team. Figures are based on Minimax’s yearly token plan pricing with the 25% yearly discount applied.
Core Concept / Estimates
25 developers, medium daily usage
| Model | Monthly Cost (Yearly Plan) |
|---|---|
| Minimax-Text-01 (1M context) | ~$150–400 |
| abab6.5t (cheap, smaller context) | ~$15–40 |
LiteLLM layer cost
| Tier | Monthly Cost |
|---|---|
| OSS (self-hosted) | Free |
| Enterprise (SSO, audit logs) | ~$50–100 |
Key Options / Variants
- Yearly Medium tier — 25% discount, sweet spot for team size; doesn’t require a huge upfront commitment
- abab6.5t — significantly cheaper; consider it if your use case doesn’t require Minimax-Text-01’s 1M context window
- Per-seat vs. pooled — a single LiteLLM proxy with one Minimax API key serves all developers; billing is by tokens, not seats
Gotchas
- Yearly token plans pre-purchase tokens at a discount. If your monthly burn rate is lower than the committed volume, you pay for unused tokens. Verify actual usage before committing.
- Token prices on the Minimax platform page (
platform.minimax.io/docs/guides/pricing-token-plan) are for yearly plans — monthly rates are higher. - Budget caps: set
max_budgetper user in LiteLLM config to prevent runaway spend from a single developer.
Updates — 2026-05-18
Ultra-Highspeed Token Plan (Individual / Small Team)
Evaluated for internal AI Workflow development and testing via Claude Code:
| Field | Value |
|---|---|
| Package | Ultra-Highspeed |
| Model | M2.7-highspeed |
| Throughput | 30,000 calls / 5 hours |
| Annual Price | $1,500 USD (~48,449 THB at May 2026 BoT rate) |
Use case: heavy AI Workflow testing where Claude Code token burn is the bottleneck. Not intended for pooled team use — see team estimates above for that.
→ See also minimax-security-privacy for enterprise data-handling posture.
Source
Conversation “Minimax and LiteLLM cost estimation for development team” — 2026-05-13 (BOL - Strategy project) Conversation “ขออนุมัติซื้อ MINIMAX Token Plan สำหรับพัฒนา AI” — 2026-05-18
Updates — 2026-05-22
Claude Code Team Premium vs MiniMax Ultra-Highspeed (Solo/Small Team)
Side-by-side comparison for a single developer or small team:
| Dimension | Claude Code Team Premium | MiniMax Ultra-Highspeed |
|---|---|---|
| Monthly price | 500/mo minimum (5 seats) | 150/mo (monthly) |
| Minimum spend | $500/month (5 seats) | $150/month (1 person) |
| Model | Sonnet 4.6 / Opus 4.6 | M2.7-highspeed |
| Context window | 1,000,000 tokens | 204,800 tokens |
| Quota unit | Tokens (~275K/5hr, third-party estimate — not official) | 30,000 requests/5hr (official) |
| Overflow | ✅ Pay API rate when over | ❌ Not documented |
| Admin/SSO | ✅ Central billing, SSO | ❌ None |
| Data privacy | ✅ No-training by default | ⚠️ No explicit “won’t train” clause (see minimax-security-privacy) |
Decision rule: If team < 5 people, MiniMax is significantly cheaper. If team ≥ 5 and context window or admin controls matter, Claude Code Team wins. Claude Code Team token quotas are not officially published — treat any third-party estimates as approximations only.
Source: Conversation “Minimax” — 2026-05-22