Breaking

MiniMax M3 ranks 7th on Code Arena Frontend with 1,531 points

June 14, 2026 at 02:22 EDT

Foundation Models
Software Dev & Coding
Open Source

Chinese AI company MiniMax has released its new model MiniMax M3, which scored 1,531 Elo to take 7th place in Arena.ai's Code Arena: Frontend and shifted the price-performance Pareto frontier in its class. Released around June 1, 2026, it is priced at $0.60 per million input tokens and $2.40 per million output tokens, drawing ratings on par with the top-tier GLM-5.1. It also placed 14th in Document Arena, with strong price efficiency standing out in both cases.

May 31 – Jun 1, 2026 · MiniMax

MiniMax M3 shifts the price–performance frontier in frontier coding

The new open-weight model takes 7th place (1531) in Code Arena: Frontend at a fraction of the usual cost — pairing 1M-token context, native multimodality, and a proprietary sparse-attention engine that runs an order of magnitude faster than M2.

7th

Code Arena: Frontend
score 1531

tokens context
512K guaranteed

$0.60/$2.40

in / out per M tokens
$0.30/$1.20 first week

MSA sparse attention vs M2

KV-block selection replaces full attention — making long-context agentic work economically practical.

1×

M2 baseline

9.7×

faster prefill

15.6×

faster decode

Vendor-reported benchmarks

SWE-Bench Pro59%

Terminal-Bench 2.166%

KernelBench Hard28.8%

Figures as cited by the vendor; independent verification limited.

Praise

1M context at a usable price tier
Frontier-class capability, cheap and open-weight
Complex frontend, game & 3D builds via MiniMax Code

Caution

Mostly vendor benchmarks, limited verification
Still breaks down on abstract reasoning
Parameter count unknown until weight release
Post-promo price sustainability unproven

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.