GLM-5.3 is live, the strongest open coding model in the world. Try it now
Sign Up
minimax

minimax-m3

Open-weight model built for coding, long-context agent work and multimodal input.

minimax 1M context
Get API Key View Pricing

Value rank

#9 of 18

intelligence per dollar in our catalogue

Context rank

#1 tied

1M token window

Agentic strength

65%

Terminal-Bench v2.1 task completion

Benchmarks

Independent scores by Artificial Analysis, compared with the strongest models in our catalogue.

Intelligence index

qwen3.8-2.4t-a95b
45.4
glm-5.3
44.9
kimi-k3
43.8
glm-5.3-flash
41.9
minimax-m3
29.6

Coding index

qwen3.8-2.4t-a95b
76.2
kimi-k3
76.2
glm-5.3
74.8
glm-5.3-flash
71.5
minimax-m3
58.6

Agentic (Terminal-Bench v2.1)

qwen3.8-2.4t-a95b
88.8%
kimi-k3
85%
glm-5.3-flash
84.3%
glm-5.3
83.9%
minimax-m3
65.2%

Long context

kimi-k3
88.7%
minimax-m3
83%
qwen3.8-2.4t-a95b
80.3%
glm-5.3-flash
80%
glm-5.3
79.7%

Best for

Where this model earns its keep.

Long-document analysis Coding and engineering Vision and image-aware tasks Prompt-cached workloads

The numbers

Pricing is live from our platform. Prices per 1M tokens, zero data retention on every request.

Input price$0.40
Cache read price$0.10
Output price$2.00
Context window1M tokens
Intelligence / coding index29.6 / 58.6
Agentic: Terminal-Bench v2.1 / tau265% / 89%
Long-context reasoning83%
GPQA / MMLU-Pro93% / -

Or consider

Close alternatives in the catalogue.

glm-5

Intelligence 27.9 · agentic -

kimi-k2.6

Intelligence 27.5 · agentic 66%

glm-5-turbo

Intelligence 26.6 · agentic -

Quick start

OpenAI-compatible. Switch in one line.

# pip install openai
client = OpenAI(base_url="https://api.tensorx.ai/v1", api_key="tsx-...")
r = client.chat.completions.create(
    model="minimax/minimax-m3",
    messages=[{"role": "user", "content": "Hello"}],
)

Benchmark data from the Artificial Analysis Intelligence Index, measured independently. Pricing live from the TensorX platform. All inference on EU-sovereign infrastructure with zero data retention.