open-source

Llama 4 Maverick 17b 128e Instruct vs Deepseek V3.2

Side-by-side pricing, context window, features, and per-use-case verdict. Data sourced from LiteLLM, updated regularly.

Specifications

Feature Llama 4 Maverick 17b 128e Instruct Deepseek V3.2
Provider groq deepseek
Model ID groq/meta-llama/llama-4-maverick-17b-128e-instruct deepseek/deepseek-v3.2
Input / 1M tokens $0.20 (29% cheaper) $0.28
Output / 1M tokens $0.60 $0.40 (33% cheaper)
Cached input / 1M
Context window 131kk tokens 164kk tokens
Max output 8kk tokens 164kk tokens
Vision
Function calling
Prompt caching
Web search
JSON schema

Monthly cost comparison

No cache discount applied — pure token costs at three usage scales.

Scale Llama 4 Maverick 17b 128e Instruct Deepseek V3.2 Winner
Small
1M in / 0.2M out
$0.32 $0.36 Llama 4 Maverick 17b 128e Instruct saves 11%
Medium
10M in / 2M out
$3.20 $3.60 Llama 4 Maverick 17b 128e Instruct saves 11%
Large
100M in / 20M out
$32.00 $36.00 Llama 4 Maverick 17b 128e Instruct saves 11%

Category winners

Cheapest input
Llama 4 Maverick 17b 128e Instruct
$0.20/M · 29% less than the other
Cheapest output
Deepseek V3.2
$0.40/M · 33% less than the other
Larger context
Deepseek V3.2
164kk tokens · 25% larger window
More features
Llama 4 Maverick 17b 128e Instruct
3 / 5 capabilities

Better for…

Input-heavy workloads (long documents, RAG)
$0.20/M input — 29% cheaper on input tokens
Llama 4 Maverick 17b 128e Instruct
Short interactions & high-throughput APIs
$0.40/M output — 33% cheaper on output at scale
Deepseek V3.2
Long-document processing & large codebases
164k context window vs 131k — 25% larger context
Deepseek V3.2
Agentic workflows & tool use
Supports 3 of 5 tracked capabilities (vision, function calling, caching, web search, JSON schema)
Llama 4 Maverick 17b 128e Instruct

Deep-dive on each model

Pricing data sourced from LiteLLM and refreshed regularly. Always verify with each provider's official pricing page before making business decisions.