Editorial illustration for MiniMax-M3 launches, beats GPT-5.5 and Gemini 3.1 Pro on benchmarks, costs 5‑10%
MiniMax-M3 launches, beats GPT-5.5 and Gemini 3.1 Pro on...
MiniMax just proved you don't need a trillion dollars to build a smart model. Their new M3 beats OpenAI and Google's flagship offerings on several key tests while costing pennies on the dollar. It is the clearest signal yet that raw performance is becoming a commodity, and that the real competition is over price and access.
On paper, the results are stark. M3 outscored GPT-5.5 and Gemini 3.1 Pro on major benchmarks. It costs between five and ten percent of what those models cost to run.
This isn't a minor improvement. It is a financial sledgehammer aimed at the foundation of the current AI economy, where capability has been locked behind prohibitively expensive API gates.
But the full picture is messier. The very hardest tasks, the ones requiring deep reasoning and complex system interaction, are still dominated by Anthropic's closed-source Opus 4.8. M3 scores a respectable 59% on the SWE-Bench Pro coding test.
Opus hits 69.2%. The gap is real on terminal commands and automated GUI work too. For the absolute peak of performance, you still pay a premium to the incumbents.
The more interesting fight is happening in the open-weight arena. Here, M3 goes head-to-head with the colossal DeepSeek-V4 Pro Max, a model with 1.6 trillion parameters. They are nearly neck and neck.
M3 has a slight edge in code synthesis. DeepSeek is a bit better at terminal execution. On web browsing and tool use, they are effectively tied.
This parity is the story. M3 achieves it not by being bigger, but by being more efficient. It uses a sparse attention mechanism to avoid activating its entire parameter set for every query.
The result is competitive power without the astronomical compute bill. It is a model built to run locally, to be owned, not rented.
In the domain of pure code modification on SWE-Bench Pro, M3's 59.0% score drops behind Opus 4.8's leading 69.2% threshold. A similar performance delta manifests in automated system environments via Terminal-Bench 2.1; while M3's 66.0% terminal execution score effectively runs neck-and-neck with the previous-generation Opus 4.7 baseline of 66.1%, it trails the upgraded Opus 4.8 architecture, which achieves 74.6%. Furthermore, evaluations tracking continuous GUI interaction on the OSWorld-Verified sandbox place M3's automated computer use at 70.0%, compared to a higher 83.4% validation rate secured by Opus 4.8.
These standardized evaluations illustrate the structural trade-offs currently defining the ecosystem: closed-source systems like Opus 4.8 maintain absolute margin leads on hyper-complex reasoning vectors, yet M3 delivers a highly capable baseline of local, tier-one automated operation without the compounding premium of closed-door API subscription fees. When positioned alongside the heavy-duty inference metrics of the newly minted, fellow open weights model DeepSeek-V4 Pro Max, M3 holds its ground across core agentic categories while asserting narrow advantages in specialized code synthesis. On the software engineering matrix of SWE-Bench Pro, M3's 59.0% resolution efficiency edges past DeepSeek-V4 Pro Max's score of 55.4%.
However, the competitive friction tightens in command-line environments; under Terminal Bench evaluations, DeepSeek-V4 Pro Max pulls slightly ahead with a 67.9% execution accuracy over M3's 66.0% mark. In web orchestration and open-world browsing simulations, the two architectures reach a virtual statistical parity, with M3 registering an 83.5% on BrowseComp compared to DeepSeek's 83.4%. Similarly, on the MCP Atlas tool-use framework, M3 secures a narrow lead at 74.2% against DeepSeek's 73.6%.This close alignment demonstrates that while DeepSeek handles a massive 1.6-trillion total parameter footprint with specialized high-effort reasoning modes, MiniMax's block-filtered sparse attention mechanism yields directly competitive execution efficiencies without requiring extensive parameter activation scaling.
The benchmark wars have always been a bit of a circus. But these numbers point to something concrete. A viable alternative is forming.
It is not about who has the single best score on one test. It is about who can deliver 95% of the capability for 5% of the cost, in a box you control. MiniMax-M3 is that alternative.
It makes the closed, expensive models look like luxury goods. For most practical work, a luxury good is a waste of money. The race is no longer just to the top of the leaderboard.
It is to the bottom of the invoice.
Related Reading
Browse all OpenAI news →
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
Related Topic OpenAI hires 630 ex-Meta staff as ChatGPT memory may turn data into ads
Business & Startups
Related Topic Sen. Warren urges Trump admin to expand AI credit after OpenAI letter
Business & Startups
From Archives Turing Award winner Richard Sutton: Pure generative AI cannot do real science
From Archives Google I/O 2026 Showcases Gemini‑Powered Infinite Scaler and Code Countdown
From Archives OpenAI begins with infrastructure robots, envisions personal robots for all
Business & Startups
From Archives OpenAI gives free life‑sciences AI model to aid government pandemic prep
Research & Benchmarks
Common Questions Answered
How does MiniMax-M3 compare in cost to GPT-5.5 and Gemini 3.1 Pro?
MiniMax-M3 costs only 5-10% of what OpenAI and Google's flagship models charge while delivering comparable performance on key benchmarks. This dramatic cost difference demonstrates that raw AI performance is becoming a commodity, with pricing and accessibility becoming the primary competitive factors rather than pure capability.
What does MiniMax-M3's benchmark performance reveal about the AI market?
MiniMax-M3 beating GPT-5.5 and Gemini 3.1 Pro on several key tests signals that viable alternatives to expensive closed models are emerging. The results show that the real competition is shifting from achieving the single best score to delivering 95% of the capability for 5% of the cost in a controllable package.
Why does the article describe closed, expensive AI models as luxury goods?
The article argues that when a model like MiniMax-M3 can deliver nearly equivalent performance at a fraction of the cost, premium-priced models become unnecessary for most practical work. For typical applications, paying significantly more for marginal performance improvements represents wasteful spending, making expensive models economically inefficient for most users.
What is the significance of MiniMax proving you don't need massive funding to build competitive AI models?
MiniMax-M3's success demonstrates that trillion-dollar budgets are not prerequisites for creating high-performing AI models that can compete with industry giants. This breakthrough suggests the AI development landscape is democratizing, with efficiency and smart engineering potentially mattering more than raw financial resources.
