Model

abab6.5

CLOSED

MiniMax · abab family · released Apr 17, 2024

MiniMax's first trillion-parameter MoE LLM, targeting GPT-4/Claude-3-class performance at launch; the abab6.5s variant trades some quality for much higher throughput at the same 200K context.

ReasoningCodingVisionFunction callingTool useAgentic

880.4

Elo · rank #242

Parameters

Undisclosed

Active params

Undisclosed

Context

200K tokens

Architecture

Mixture-of-Experts, trillion-scale total params (exact count/expert configuration undisclosed)

License

Proprietary (MiniMax API terms)

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

BIG-Bench HardReasoning83.1%#23

best: ERNIE 4.5 300B-A47B · 94.3%

GSM8KMath91.7%#33

best: Llama 3.1 405B · 96.8%

HumanEvalCoding78.0%#59

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning78.9%#93

best: Gemma 4 26B A4B · 98.5%

MATH-500Math51.3%#127

best: GPT-5 · 99.4%

MMLUKnowledge78.7%#101

best: OpenAI o3 · 92.9%

MT-BenchHuman preference8.77#16

best: Hunyuan-Large (A52B) · 9.4

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

—

Output / M tok

—

abab family

Elo progression across releases

abab6.5Apr 2024880.4

Resources

DOCSplatform.minimax.io/docs APIAvailable via MiniMax API

API price — · each benchmark row carries its own source badge (see methodology)