Llama 4 Maverick vs Mistral Large 3 2512

Pricing math

Provider list price: Maverick charges $0.15/M input and $0.60/M output; Mistral Large 3 charges $0.50/M input and $1.50/M output. On OpenKey that becomes Maverick at $0.1545/M in and $0.618/M out ($0.15 x 1.03, $0.60 x 1.03), and Mistral Large 3 at $0.515/M in and $1.545/M out ($0.50 x 1.03, $1.50 x 1.03).

For a 10M-input / 2M-output workload, Maverick costs $2.70 total versus Mistral Large 3's $8.00 — Maverick is roughly a third of the price (input price ratio 0.3). If you're running high-volume pipelines where quality differences don't matter much, that gap compounds fast. Mistral Large 3 also offers cache reads at $0.05/M, which Maverick doesn't support, so repeated-context workloads on Mistral can claw back some of that gap.

Coding and agentic performance

Mistral Large 3 leads on every measured axis here. Artificial Analysis coding index: 20.1 vs Maverick's 16.3. Agentic index: 5.5 vs 1.3 — a real gap for tool-use-heavy workflows. Design Arena's codecategories benchmark backs this up: Mistral Large 3 scores 1191 elo (rank 56, 47.6% win rate) versus Maverick's 929 elo (rank 102, 35.8% win rate).

The pattern holds across gamedev (Mistral 1146 elo/rank 65 vs Maverick 903 elo/rank 103), dataviz (1180/rank 55 vs 926/rank 100), uicomponent (1157/rank 59 vs 955/rank 94), and website (1205/rank 53 vs 914/rank 105). If your workload is code generation or agent orchestration, Mistral Large 3 is the stronger pick by a wide margin on every category both models were tested in.

Context and long-document work

Maverick supports a 1,048,576-token context window — 4x Mistral Large 3's 262,144 tokens (context ratio 4.0). For jobs that mean ingesting large codebases, long transcripts, or big document sets in a single call, Maverick's window gives you more headroom without chunking. Maverick's max completion is capped at 16,384 tokens; Mistral Large 3 doesn't publish a max completion limit in this data. Both accept image input; Mistral Large 3 also accepts file input directly, which Maverick's spec doesn't list. If the job is fundamentally about fitting more into one context window at low cost, Maverick has the clear edge here.

When to pick each

Pick Llama 4 Maverick when you're processing high token volumes, need the largest context window available between the two, or are cost-constrained and can tolerate lower coding/agentic scores. Pick Mistral Large 3 2512 when the task is coding, agent workflows, or anything where the Design Arena and Artificial Analysis gaps translate to fewer retries and better first-pass output. Mistral Large 3 is also Apache 2.0 licensed and newer (created December 2025 vs Maverick's April 2025), which matters if license terms or recency of training data are decision factors for you.

Questions

Which model is cheaper for a typical workload?

Llama 4 Maverick. A 10M-input/2M-output job costs $2.70 on Maverick versus $8.00 on Mistral Large 3 2512 — Maverick's input price is roughly 0.3x Mistral's ($0.1545/M vs $0.515/M on OpenKey after the 3% fee).

Which model has the bigger context window?

Llama 4 Maverick, at 1,048,576 tokens versus Mistral Large 3 2512's 262,144 tokens — a 4x ratio. That matters for long-document or large-codebase tasks that need to fit in a single call.

Which model is better at coding?

Mistral Large 3 2512, with a coding index of 20.1 versus Maverick's 16.3, and a 1191 elo score on Design Arena's codecategories benchmark (rank 56) versus Maverick's 929 elo (rank 102).

Is one model newer than the other?

Yes. Mistral Large 3 2512 was created December 1, 2025; Llama 4 Maverick was created April 5, 2025 — about eight months earlier. Mistral Large 3 is also released under Apache 2.0.