Gemma 4 31B vs Llama 3.3 70B Instruct vs Grok 4.20

Gemma 4 31B at $0.115 in and $0.391 out, Llama 3.3 70B Instruct at $0.115 in and $0.368 out, and Grok 4.20 at $1.44 in and $2.88 out — per million tokens, from the same wallet.

Gemma 4 31B

Google

0.09¢

for a client proposal

$0.115 in · $0.391 out / 1M

262K context · released 4 months ago

Full page →

Llama 3.3 70B Instruct

At a glance

Cheapest input: Gemma 4 31B at $0.115 per million tokens.
Cheapest output: Llama 3.3 70B Instruct at $0.368 per million tokens.
Largest context: Grok 4.20 at 2M tokens — about 2,900 pages.

Cost per task

Estimated totals at the rates above. Solid bar is input, lighter bar is output.

Quick code fix

2K in · 500 out

Gemma 4 31B

0.04¢

Llama 3.3 70B Instruct

0.04¢

Grok 4.20

0.43¢

Draft a client proposal from notes

5K in · 800 out

Gemma 4 31B

0.09¢

Llama 3.3 70B Instruct

0.09¢

Grok 4.20

0.95¢

Context-heavy session (Alyph workspace)

155K in · 4K out

Gemma 4 31B

1.9¢

Llama 3.3 70B Instruct

needs 155K ctx

Grok 4.20

$0.23

Analyze a 50-page legal contract

35K in · 1K out

Gemma 4 31B

0.44¢

Llama 3.3 70B Instruct

0.44¢

Grok 4.20

$0.05

Large architecture refactor

250K in · 3K out

Gemma 4 31B

3¢

Llama 3.3 70B Instruct

needs 250K ctx

Grok 4.20

$0.74

Long-context rate applied at this input size.

Solid part of each bar is input tokens, the lighter part is output. Models that don’t fit a task are marked instead of priced.

Cost vs Token Amount

Drag the slider to adjust the split between input and output workload.

80% Input20% Output

Gemma 4 31B

Llama 3.3 70B Instruct

Grok 4.20

Adjust the ratio slider to change how output tokens influence the final price for equivalent workloads.

Pricing

Gemma 4 31B

Llama 3.3 70B Instruct

Grok 4.20

Input / 1M tokens

$0.115

$1.44

Output / 1M tokens

$0.391

$0.368

$2.88

Cached input / 1M

$0.115

—

$0.230

Web search

—

0.57¢ / search

Specs

Gemma 4 31B

Llama 3.3 70B Instruct

Grok 4.20

Provider

Google

Try it with real numbers

2K tokens in · 500 tokens out.

Gemma 4 31B

Google

0.04¢

for this task

54% input · 46% output

Full pricing →

Llama 3.3 70B Instruct

Which should you pick?

For long documents, Grok 4.20 has the largest context window here — 2M tokens.

On price, Llama 3.3 70B Instruct is the cheapest of the three on a typical task, at about 0.09¢.

For images or files, Gemma 4 31B and Grok 4.20 can read them directly; Llama 3.3 70B Instruct is text-only.

If you want to watch the model think, Gemma 4 31B and Grok 4.20 show reasoning; the other answers directly.

Or don’t pick. Send the identical prompt to all three, read the answers side by side, and keep the winner. How to run a fair bake-off →

Gemma 4 31B pricing →Llama 3.3 70B Instruct pricing →Grok 4.20 pricing →Try other combinations →

Questions, answered

Which is cheapest — Gemma 4 31B, Llama 3.3 70B Instruct, or Grok 4.20?

On a typical client proposal (5K tokens in, 800 out): Llama 3.3 70B Instruct at 0.09¢; Gemma 4 31B at 0.09¢; Grok 4.20 at 0.95¢. Long prompts can change the order when long-context rates apply.

Which has the largest context window?

Grok 4.20 — 2M tokens, about 2,900 pages.

Can I run Gemma 4 31B, Llama 3.3 70B Instruct, and Grok 4.20 side by side on Alyph?

Yes — that is what Alyph is built for: the identical prompt with identical context to all 3, answers rendered side by side, billed from one wallet. This exact combination costs about 1.1¢ for a typical task.

Ask all three at once

Identical prompt, identical context, answers side by side — about 1.1¢ for a typical client proposal. Free to start, with $5 of credit.