DeepSeek V4 Pro vs MiniMax M3: The Same Open-Weights Game at Five Times the Price

DeepSeek V4 Pro and MiniMax M3 are both open-weights reasoning models with a million-token context, and they are both cheap enough that the word "budget" no longer means "weak." But their launches could not look more different: MiniMax M3 (Mini​Max's 428B-parameter model, released June 2026) went out with a community license and a $0.30/$1.20 per million token price that undercut the entire field, while DeepSeek V4 Pro (released August 13, 2026, at $0.66/$1.98 off-peak, $1.32/$3.96 peak) arrived six weeks later with MIT-licensed weights and a slightly higher price. The twist is what happened to the more expensive one: between September 8 and 11, Deep​Seek first scheduled V4 Pro for retirement, then delayed it, then cancelled the retirement entirely, committing to keep serving it at unchanged billing. So the matchup that a month ago looked like "cheap and cheaper" now has a durability story on the one side that no one expected — and that makes it worth re-reading the whole thing.

Independent measurements below are from Artificial Analysis. Vendor-reported figures are labeled as such.

Two budget open-weights flagships, one index point apart

On the Artificial Analysis Intelligence Index, MiniMax M3 scores 30, ranked #16 in its class, and DeepSeek V4 Pro scores 36, ranked #7. A six-point gap between models that cost almost the same per token. That is the headline: the more expensive model is also the measurably smarter one — but only by six points, and only on the text-only axis where V4 Pro lives. MiniMax M3 takes text, image, and video input; V4 Pro is text-only. If multimodal input is a requirement, the comparison ends there — MiniMax M3 is the only one of the two that reads video.

• Intelligence Index — V4 Pro 36 (#7/113) vs MiniMax M3 30 (#16/113)

• Price — MiniMax M3 $0.30/$1.20 per 1M vs V4 Pro $0.66/$1.98 off-peak ($1.32/$3.96 peak)

• Speed — MiniMax M3 ~103 tokens/sec vs V4 Pro ~81 tokens/sec

• Context — both 1M tokens

• Parameters — MiniMax M3 428B total / 23B active vs V4 Pro 1.6T total / 49B active

• Input — MiniMax M3 text/image/video vs V4 Pro text-only

• License — Mini​Max Community License vs V4 Pro MIT

The price race is closer than the sticker suggests

Listed head to head, MiniMax M3 is 2.2x cheaper on input and 1.65x cheaper on output than V4 Pro's off-peak rate. But the sticker prices hide the cache and peak mechanics. V4 Pro's off-peak window cuts its already-low prices in half, and its cache-read discount runs to about 97% — around $0.02 per million cached tokens off-peak. MiniMax M3's cache pricing is less aggressive. For a workload that reuses a large prefix — a big system prompt, a long document, a retrieval corpus — the effective per-token cost of V4 Pro in its off-peak cache window can undercut MiniMax M3's headline price, which is a surprising inversion of the sticker ranking. On fresh-token workloads, MiniMax M3 is cheaper; on cache-heavy workloads, V4 Pro can win.

Speed tells a similar story in reverse. MiniMax M3 outputs about 103 tokens per second to V4 Pro's 81 — the fastest of any model at its price tier, and faster than several models that cost five times as much. The speed edge is real but not enormous; the price edge matters more, and it is workload-dependent.

License and durability: where September changed the answer

Both models are open weights, but the licenses differ. MiniMax M3 ships under the MINI​MAX COMMUNITY LICENSE — permissive in spirit, but with its own terms. V4 Pro is MIT — about as unrestricted as an open-weights model gets, covering commercial use, fine-tuning, redistribution, and closed modifications. For enterprise adoption the MIT license is the cleaner story, and the gap shows up the moment your legal team reads the terms.

The durability question, which was a real one a month ago, is the September news. Deep​Seek spent the first half of September announcing V4 Pro's retirement for September 14, delaying it, and then — on the 11th — cancelling it, with a public commitment to keep serving the model at unchanged billing and to give notice if that changes. MiniMax M3 has no such drama because none was expected: it is a young flagship with no retirement in view. But the effect is the same: both sides of this matchup now have a clear roadmap, which is more than could be said for one of them two weeks ago.

Who should pick which

If your workload includes image or video input, the decision is made: MiniMax M3 is the only one of the two that reads them, at the lowest price in its class. If your workload is text-only and cache-heavy — agents with large system prompts, long-document RAG, summarization of big corpora — V4 Pro's off-peak cache window and MIT license make it the better cost-per-answer pick, especially since it is now guaranteed to stay. If your workload is text-only and latency-sensitive, MiniMax M3's 103 tokens/sec and $0.30/$1.20 make it the fastest cheap option on the board.

Both models are available through OrcaRouter on one key, with provider list prices passed through at zero markup — so the numbers above are the numbers you pay — and automatic failover between providers is the practical answer to the one thing this matchup can't settle in advance: which model's provider will hold up under your actual load. Routing the multimodal traffic to MiniMax M3 and the long-context text work to DeepSeek V4 Pro is a two-line routing-DSL change, and the failover means a provider wobble on either side is a retry, not an incident.

The verdict

This is the closest matchup in the DeepSeek V4 Pro lineup, because both models are cheap, both are open weights, and both are fast. MiniMax M3 wins on raw price, speed, and multimodal input. DeepSeek V4 Pro wins on intelligence, license cleanliness, and — since September 11 — durability certainty. The one-line answer: multimodal or latency-critical text goes to MiniMax M3; cache-heavy, text-only, or legally-sensitive workloads go to DeepSeek V4 Pro.

Route the multimodal traffic to MiniMax M3 and the long-context text work to DeepSeek V4 Pro: MiniMax M3 a two-line routing-DSL change.