MiniMax M3 vs Grok 4.3
MiniMax M3 (MiniMax, released 2026-05-31) and Grok 4.3 (xAI, released 2026-04-30) both ship 1M-token-class context windows and target agentic and coding work, but they split hard on price and on which benchmark arena they actually show up in. M3 is multimodal (text, image, video in), Grok 4.3 adds file input instead of video. Both run on OpenKey under one API key with a flat 3% fee on top of provider list price.
Spec vs spec
| Spec | MiniMax M3 | Grok 4.3 |
|---|---|---|
| Context window | 1.0M | 1M |
| Max output | 512K | — |
| Input modalities | text, image, video | text, image, file |
| Output modalities | text | text |
| Released | May 31, 2026 | Apr 30, 2026 |
| 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%
x-ai/grok-4.3
Input · 1M tokens
$1.25 + 3%$1.29
Output · 1M tokens
$2.50 + 3%$2.58
Cache read · 1M tokens
$0.200 + 3%$0.206
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%
x-ai/grok-4.3
$18.03
$17.50 provider + 3%
Benchmarks
Design Arena categories where both models have results. Higher Elo and lower rank win.
| MiniMax M3 | Grok 4.3 | |||
|---|---|---|---|---|
| Category | Elo | Rank | Elo | Rank |
| 3D | 1306 | #16 | 1202 | #43 |
| androidnative | 990 | #27 | 1074 | #23 |
| ASCII art | 1219 | #15 | 1189 | #27 |
| Code | 1306 | #13 | 1243 | #31 |
| Data viz | 1295 | #11 | 1234 | #35 |
| Game dev | 1287 | #20 | 1242 | #34 |
| SVG | 1250 | #13 | 1140 | #44 |
| UI components | 1291 | #18 | 1250 | #31 |
| Websites | 1304 | #11 | 1244 | #31 |
Head-to-head preference voting. How we filter and rank
Pricing: the gap is real
MiniMax M3 costs $0.30/M input and $1.20/M output at the provider; on OpenKey that's $0.309/M and $1.236/M (provider price × 1.03). Grok 4.3 runs $1.25/M input and $2.50/M output, or $1.2875/M and $2.575/M on OpenKey. That's an input price ratio of 0.24 — M3 is about a quarter the cost per input token. Run the math on a real workload: 10M input tokens + 2M output tokens costs $5.40 on MiniMax M3 versus $17.50 on Grok 4.3. If you're doing high-volume coding or agent loops where token count adds up fast, that difference compounds daily.
Design Arena: two different competitions
MiniMax M3 doesn't appear in Design Arena's agents arena for mobileapps, webapps, fullstack, or godotgamedev — Grok 4.3 does, and wins those categories outright by default since M3 has no entry there. M3's only agents-arena listing is androidnative, where it ranks 27th (elo 990, 20.6% win rate) — worse than Grok 4.3's androidnative rank of 23rd (elo 1074, 28.6% win rate). If your workload is building mobile apps, full-stack web apps, or Godot games through an agent, Grok 4.3 is simply the only one of the two with a track record there.
Model arena: M3 wins where both compete
In the shared model-arena categories, MiniMax M3 outranks Grok 4.3 across the board: codecategories (M3 rank 13 vs Grok rank 31), svg (13 vs 44), uicomponent (18 vs 31), website (11 vs 31), dataviz (11 vs 35), 3d (16 vs 43), gamedev (20 vs 34), asciiart (15 vs 27). Artificial Analysis backs this up — M3 scores 58.6 on coding index versus Grok's 42.2, and 44.4 on the intelligence index versus 37.6. For anything scored in the general model arena, not the agents arena, M3 is the stronger pick.
Context and modality
MiniMax M3 has a 1,048,576-token context window and can output up to 512,000 tokens — the largest completion cap in this pair. Grok 4.3 offers 1,000,000 tokens of context (a context ratio of 1.05, effectively a tie) with no stated max-completion limit in the catalog. Modality differs: M3 accepts text, image, and video; Grok 4.3 accepts text, image, and file input, plus it exposes selectable reasoning effort (high, medium, low, none — default low), which M3 does not expose as a tunable parameter.
When to pick each
Pick MiniMax M3 for coding-heavy pipelines, SVG/UI/website generation, video-input tasks, or any workload where cost per token matters — it's cheaper and ranks higher everywhere the two overlap in the model arena. Pick Grok 4.3 for agentic app-building tasks it has a track record in — mobile apps, full-stack web apps, Godot game dev, or PPTX/HTML slide generation — or when you need file-input support and adjustable reasoning effort. If your task isn't in Grok's agents-arena list at all, there's no benchmark reason to pay the premium.
Which model for which job
| Use case | Pick | Why |
|---|---|---|
| High-volume coding agent | MiniMax M3 | Coding index 58.6 vs 42.2, and costs $5.40 vs $17.50 on the 10M-in/2M-out workload |
| Mobile app agent build | Grok 4.3 | Ranks 26th in mobileapps agents arena (elo 1146); MiniMax M3 has no entry in that category |
| Full-stack web app agent | Grok 4.3 | Ranks 26th in fullstack agents arena (elo 1072); MiniMax M3 isn't listed there |
| SVG / UI component generation | MiniMax M3 | Ranks 13th on svg and 18th on uicomponent, both well ahead of Grok's 44th and 31st |
| Video-input analysis | MiniMax M3 | Only one of the two that accepts video as an input modality |
| Godot game dev agent | Grok 4.3 | Ranks 16th in godotgamedev agents arena (elo 1134); no MiniMax M3 entry exists |
Questions
- Which model is cheaper for a typical 10M-input, 2M-output job?
- MiniMax M3 costs $5.40 on OpenKey for that workload; Grok 4.3 costs $17.50. The input price ratio is 0.24, meaning M3's input tokens run about a quarter of Grok 4.3's price.
- Does Grok 4.3 actually beat MiniMax M3 on benchmarks?
- Only in categories MiniMax M3 doesn't appear in at all — mobileapps (rank 26), fullstack (rank 26), webapps (rank 14), godotgamedev (rank 16). Where both are ranked in the model arena, M3 wins every listed category, including codecategories (13 vs 31) and website (11 vs 31).
- Which has the bigger context window?
- MiniMax M3 has 1,048,576 tokens versus Grok 4.3's 1,000,000 — a context ratio of 1.05, essentially a tie. M3 also allows up to 512,000 output tokens; Grok 4.3 has no stated max-completion cap in the catalog.
- Can I use both through one API key?
- Yes — both MiniMax M3 and Grok 4.3 run on OpenKey with a single key and a flat 3% fee on provider list price, so OpenKey pricing works out to $0.309/$1.236 per M tokens for M3 and $1.2875/$2.575 per M tokens for Grok 4.3.