Skip to content
LIVE
OPUS 4.8$5 / $25per Mtok
GPT-5.5$5 / $30per Mtok
GEMINI 2.5 PRO$1.25 / $10per Mtok
SONNET 4.6$3 / $15per Mtok
SWE-BENCHleader GPT-5.568.7%
MMLU-PROleader GPT-5.594.2
GPQAleader GPT-5.578.3
AFTAv1.0 whitepaper live at /whitepaper
OPUS 4.8$5 / $25per Mtok
GPT-5.5$5 / $30per Mtok
GEMINI 2.5 PRO$1.25 / $10per Mtok
SONNET 4.6$3 / $15per Mtok
SWE-BENCHleader GPT-5.568.7%
MMLU-PROleader GPT-5.594.2
GPQAleader GPT-5.578.3
AFTAv1.0 whitepaper live at /whitepaper
All systems operational0 AI providers monitored, polled every 2 minutes
Live status

MiniMax M3 vs DeepSeek V4 Pro

China's two loudest open-weight labs now have flagships aimed at the same workload: cheap, long-context agentic coding. MiniMax M3 landed June 1, 2026 with sparse attention that cuts 1M-context compute to roughly one twentieth of its previous generation, $0.30/$1.20 pricing, and a claimed 59% on SWE-Bench Pro. DeepSeek V4 Pro, the April 2026 incumbent, counters with a verified 80.6% on SWE-bench Verified, MIT-licensed weights you can download today, and a longer record of independent benchmarking. The big caveat: M3's headline numbers were run on MiniMax's own infrastructure with agent scaffolding, and its weights are still about ten days out at launch. V4 Pro is the safer pick today; M3 is the one to watch.

Head-to-Head Specs

SpecMiniMax M3DeepSeek V4 Pro
ProviderMiniMaxDeepSeek
Input Price$0.30/1M$0.43/1M
Output Price$1.20/1M$0.87/1M
Context Window1.0M1M
Released2026-062026-04
Capabilitiestext, vision, video, code, tool-usetext, vision, code, reasoning

Category Breakdown

PricingMiniMax M3

M3 costs $0.30/$1.20 per million tokens vs V4 Pro at $0.435/$0.87; M3 is cheaper on input, V4 Pro on output

Verified benchmarksDeepSeek V4 Pro

V4 Pro has independently reproduced scores including 80.6% SWE-bench Verified; M3 launch numbers are self-reported with agent scaffolding

Long-context efficiencyMiniMax M3

MiniMax Sparse Attention cuts per-token compute at 1M context to roughly 1/20th of the prior generation

Context windowTieTie

Both ship roughly 1M token context windows

Multimodal inputMiniMax M3

M3 accepts text, image, and video input; V4 Pro covers text and vision

Weights availabilityDeepSeek V4 Pro

V4 Pro weights are on Hugging Face under MIT today; M3 weights were still pending about ten days after launch

License clarityDeepSeek V4 Pro

V4 Pro is MIT licensed; M3's license is unconfirmed until the weights drop

Choose MiniMax M3 when:

  • Cost-sensitive agentic pipelines with massive input context
  • Browser and tool-use agents (83.5 BrowseComp claimed)
  • Video and image input alongside code
  • Workloads that can tolerate launch-window uncertainty
View MiniMax M3 details

Choose DeepSeek V4 Pro when:

  • Self-hosting today under a clear MIT license
  • Workloads that need independently verified coding performance
  • Output-heavy generation where $0.87 beats $1.20
  • Production systems that cannot wait for M3 verification
View DeepSeek V4 Pro details

Frequently Asked Questions

Which is better, MiniMax M3 or DeepSeek V4 Pro?

It depends on your use case. MiniMax M3 from MiniMax excels at cost-sensitive agentic pipelines with massive input context, while DeepSeek V4 Pro from DeepSeek is better for self-hosting today under a clear mit license. See the full comparison above for detailed benchmarks and pricing.

How much does MiniMax M3 cost compared to DeepSeek V4 Pro?

MiniMax M3 costs $0.30 input and $1.20 output per 1M tokens. DeepSeek V4 Pro costs $0.43 input and $0.87 output per 1M tokens.

What is the context window difference between MiniMax M3 and DeepSeek V4 Pro?

MiniMax M3 supports 1.0M tokens, while DeepSeek V4 Pro supports 1M tokens.

More Comparisons

Interactive Compare ToolAll ModelsFull Pricing Guide