MiniMax M3 vs Qwen3 Max
MiniMax M3 and Qwen3 Max are both non-reasoning-mandatory models aimed at coding and agentic tasks, but they land in different price and capability tiers. M3 is newer (created 2026-05-31 vs Qwen3 Max's 2025-09-23), takes text, image, and video input, and ships a 1M-token context window against Qwen3 Max's 262,144 tokens. The gap shows up directly in Design Arena scores and in what a real workload costs on OpenKey.
Spec vs spec
| Spec | MiniMax M3 | Qwen3 Max |
|---|---|---|
| Context window | 1.0M | 262K |
| Max output | 512K | 33K |
| Input modalities | text, image, video | text |
| Output modalities | text | text |
| Knowledge cutoff | — | Jun 30, 2025 |
| Released | May 31, 2026 | Sep 23, 2025 |
| Reasoning | optional | optional |
Pricing
Per 1M tokens. Provider price plus the flat 3% fee — the sum is what you pay.
minimax/minimax-m3
Input · 1M tokens
$0.300 + 3%$0.309
Output · 1M tokens
$1.20 + 3%$1.24
Cache read · 1M tokens
$0.060 + 3%$0.062
FEE — FLAT, EVERY MODEL3%
qwen/qwen3-max
Input · 1M tokens
$0.780 + 3%$0.803
Output · 1M tokens
$3.90 + 3%$4.02
Cache read · 1M tokens
$0.156 + 3%$0.161
Cache write · 1M tokens
$0.975 + 3%$1.00
FEE — FLAT, EVERY MODEL3%
One workload, priced on both
10M input + 2M output tokens at each model's price, flat 3% fee included.
minimax/minimax-m3Cheaper
$5.56
$5.40 provider + 3%
qwen/qwen3-max
$16.07
$15.60 provider + 3%
Benchmarks
Design Arena categories where both models have results. Higher Elo and lower rank win.
| MiniMax M3 | Qwen3 Max | |||
|---|---|---|---|---|
| Category | Elo | Rank | Elo | Rank |
| 3D | 1306 | #16 | 1150 | #62 |
| ASCII art | 1219 | #15 | 1175 | #32 |
| Code | 1306 | #13 | 1159 | #66 |
| Data viz | 1295 | #11 | 1149 | #64 |
| Game dev | 1287 | #20 | 1160 | #62 |
| SVG | 1250 | #13 | 1069 | #60 |
| UI components | 1291 | #18 | 1132 | #67 |
| Websites | 1304 | #11 | 1161 | #66 |
Head-to-head preference voting. How we filter and rank
Pricing math
Provider list price for MiniMax M3 is $0.30/1M input and $1.20/1M output; for Qwen3 Max it's $0.78/1M input and $3.90/1M output. On OpenKey, both get the flat 3% fee: M3 becomes $0.309/$1.236 per 1M (0.30 × 1.03, 1.20 × 1.03), Qwen3 Max becomes $0.8034/$4.017 per 1M (0.78 × 1.03, 3.90 × 1.03).
Run the numbers on a realistic 10M-input / 2M-output workload: M3 costs $5.40 total, Qwen3 Max costs $15.60 — Qwen3 Max is nearly 3x more expensive for the same job. The input price ratio alone is 0.38 (M3 is 38% of Qwen3 Max's input cost). M3 also has a cache-read price of $0.06/1M vs Qwen3 Max's $0.156/1M, so repeated-context workloads widen the gap further.
Coding and agentic performance
MiniMax M3 has published Artificial Analysis scores that Qwen3 Max doesn't: intelligence index 44.4, coding index 58.6, agentic index 35.4. On Design Arena, M3 outranks Qwen3 Max in every shared category — codecategories (rank 13 vs 66, elo 1306 vs 1159), dataviz (rank 11 vs 64, elo 1295 vs 1149), svg (rank 13 vs 60, elo 1250 vs 1069), website (rank 11 vs 66, elo 1304 vs 1161), uicomponent (rank 18 vs 67), gamedev (rank 20 vs 62), asciiart (rank 15 vs 32), and 3d (rank 16 vs 62). M3 also has an androidnative agents-arena score (elo 990, rank 27) that Qwen3 Max isn't tested on at all.
Context and long-document work
M3's 1,048,576-token context window is 4x Qwen3 Max's 262,144 tokens (the context ratio computed directly from both catalog entries). M3's max completion is also far larger — 512,000 tokens vs Qwen3 Max's 32,768 — so for tasks that need to return long generated output (full codebases, long reports, multi-file diffs) M3 has more room to work with in a single call. If your workload chunks a large codebase or a long transcript into one context, M3 avoids the extra round-trips Qwen3 Max would need.
Modality and tooling
M3 accepts text, image, and video input; Qwen3 Max is text-only in and out. M3 also exposes more supported parameters — including `reasoning`, `include_reasoning`, `min_p`, `top_k`, and `repetition_penalty` — versus Qwen3 Max's shorter list (`logprobs`, `presence_penalty`, `response_format`, `seed`, `structured_outputs`, `temperature`, `tool_choice`, `tools`, `top_logprobs`, `top_p`). Neither model has reasoning marked mandatory. If your pipeline needs to feed screenshots or video frames alongside text, Qwen3 Max simply can't take that input — M3 is the only option of the two.
When to pick each
Pick MiniMax M3 by default: it's cheaper, has a bigger context window, and wins every comparable benchmark category. Qwen3 Max's one edge is a stated knowledge cutoff (2025-06-30) — useful if you need to know exactly how current the model's training data is and M3's cutoff isn't documented. Outside that narrow case, there's no benchmark or price dimension in this data where Qwen3 Max comes out ahead. Both models run on OpenKey through one API key with a flat 3% fee over provider list price, so switching between them mid-project is a one-line change, not a new account.
Which model for which job
| Use case | Pick | Why |
|---|---|---|
| Coding agents / long-horizon tasks | MiniMax M3 | Higher coding index (58.6) and agentic index (35.4), plus top ranks across every Design Arena code-related category |
| Large codebase or long-document context | MiniMax M3 | 1,048,576-token context vs 262,144 — a 4x window advantage |
| Cost-sensitive high-volume workload | MiniMax M3 | $5.40 vs $15.60 for a 10M-input/2M-output job — under a third of the cost |
| Image or video input pipelines | MiniMax M3 | Only M3 supports image and video input modalities; Qwen3 Max is text-only |
| Time-sensitive factual work needing a known cutoff | Qwen3 Max | Documented knowledge cutoff of 2025-06-30; M3's cutoff isn't specified |
| Long single-shot generation (full reports, multi-file output) | MiniMax M3 | Max completion of 512,000 tokens vs Qwen3 Max's 32,768 |
Questions
- How much cheaper is MiniMax M3 than Qwen3 Max?
- On OpenKey, M3 runs $0.309/1M input and $1.236/1M output; Qwen3 Max runs $0.8034/1M input and $4.017/1M output. For a 10M-input/2M-output workload, that's $5.40 total for M3 versus $15.60 for Qwen3 Max — Qwen3 Max costs almost 3x more for the same job.
- Which model has a bigger context window?
- MiniMax M3 supports 1,048,576 tokens of context, four times Qwen3 Max's 262,144 tokens. M3's max output is also much larger at 512,000 tokens versus Qwen3 Max's 32,768, which matters for tasks returning long generated content.
- Does either model handle images or video?
- MiniMax M3 takes text, image, and video input and outputs text. Qwen3 Max is text-only in both directions. If your app needs to send screenshots, diagrams, or video frames to the model, M3 is the only one of the two that can process them.
- Which model wins on coding benchmarks?
- MiniMax M3 leads on every shared Design Arena coding category — codecategories rank 13 vs Qwen3 Max's rank 66, svg rank 13 vs 60, and website rank 11 vs 66. M3 also has a published Artificial Analysis coding index of 58.6, a metric not available for Qwen3 Max.