Llama vs Gemma: benchmarks and API costs

Gemma 4 31B leads SciCode at 45.49% versus 31.71%. Gemma 4 31B has the lower illustrated request cost at $0.0018; the other model costs $0.00355.

Updated 2026-10-29 · 4 sources · Methodology

Dimension	Winner	Margin	Evidence
SciCode (%)	Gemma 4 31B	13.78 pp	Llama 4 Maverick 31.71% / Gemma 4 31B 45.49%¹²
Output speed (tokens/s)	Llama 4 Maverick	14.5 tokens/s	Llama 4 Maverick 50.0 / Gemma 4 31B 35.5 tokens/s
AA time to first answer token (s)	Llama 4 Maverick	49.07 s	Llama 4 Maverick 0.97 s / Gemma 4 31B 50.04 s
Estimated request cost	Gemma 4 31B	$0.00175	10K + 1K: Llama 4 Maverick $0.00355 / Gemma 4 31B $0.0018

AA coding quality benchmarks

Metric	Llama 4 Maverick	Gemma 4 31B (Reasoning)
SciCode (%)	31.71¹	45.49²

Performance

Metric	Llama 4 Maverick	Gemma 4 31B (Reasoning)
Output speed (tokens/s)	50.0	35.5
AA time to first answer token (s)	0.97	50.04

Prices and request cost

Metric	Llama 4 Maverick · Novita AI	Gemma 4 31B · Novita AI
Input / 1M tokens	$0.27³	$0.14⁴
Output / 1M tokens	$0.85³	$0.4⁴
Context window (tokens)	1,048,576³	262,144⁴
Max output (tokens)	8,192³	131,072⁴
API documentation	Novita AI · Chat API

Cost example: 10,000 input + 1,000 billed output tokens

Llama 4 Maverick$0.00355

Gemma 4 31B$0.0018

Coding, reasoning, context and cost

Coding benchmark

AA SciCode: Llama 4 Maverick 31.71% / Gemma 4 31B 45.49%.¹²

Reasoning benchmark

The configured AA first-answer wait is 0.97 s for Llama 4 Maverick and 50.04 s for Gemma 4 31B.¹²

Context window (tokens)

Llama is the capacity choice at 1,048,576 tokens; Gemma’s endpoint stops at 262,144.¹²

Estimated request cost

The same uncached token quantities cost $0.00355 for Llama 4 Maverick and $0.0018 for Gemma 4 31B.¹²

Methodology and sources

SciCode is AA’s scientific-coding evaluation, reported as a percentage; higher is better. It is separate from the 10,000-input-token performance test and is not an in-editor Cline evaluation. Quality and performance use the configurations shown beside the model names. A current AA v5 Intelligence Index is displayed only where both models have a published value. AA performance uses the 10,000-input-token benchmark at each model’s displayed configuration; reasoning settings vary by model. Output speed is the median output rate. Answer latency measures the first answer token after reasoning, not the initial streamed reasoning chunk. For non-streamed reasoning and non-reasoning models, AA reports the same first-chunk and first-answer event. AA estimates answer latency from initial response and reasoning duration; it is not a separately sampled median. These are reference deployments, not Novita endpoint guarantees. The uncached cost example bills 10,000 input and 1,000 output tokens at each named seller’s tariff; it is not the cost of the AA benchmark. Prices, context and output limits use the cited seller sources.

Build with Novita Model APIs

Create an API key and connect your application.

Start Building Explore Models