Cheaper alternatives to Llama 3.2 3B Instruct

Meta LMArena 1109.7 $0.120/M on a balanced workload prices as of 2026-08-28

9 models are both better and cheaper, giving up nothing.

Of the 133 models carrying both an independent score and a published price at the same delivery mode, 17 score at least as high as Llama 3.2 3B Instruct and cost no more under at least one workload. 9 match or beat it on every capability we hold — context, maximum output, input modes, tool use and extended reasoning. 8 do not, and every row names what it drops.

Dominance is not a property of a model. It is a property of a model, a capability lens and a workload, and all three are named on every row. Scores are LMArena Elo, used under CC BY 4.0 from the official dataset; prices are blended per million tokens. A higher score is not a drop-in replacement.

This page is built from that comparison and nothing else. When no model is both better and cheaper than Llama 3.2 3B Instruct, it is not generated.

Where it holds

Workload	Llama 3.2 3B Instruct $/M	Better and cheaper	With a trade
Balanced	$0.120	6	6
Summarise	$0.064	6	5
Chat	$0.162	7	7
Code gen	$0.218	7	7
Agentic	$0.092	9	6

Workload mixes are defined on the methodology page. Cached input is priced at the cached rate, and reasoning tokens at the worse of the reasoning and output rates.

What you would be replacing

Intelligence	1109.7
Context window	131K
Max output	118K
Input modes	text
Tool use	no
Extended reasoning	no

A replacement has to clear every line above, not just the score. Full record for Llama 3.2 3B Instruct.

Better and cheaper, nothing given up 9

Each of these matches or beats Llama 3.2 3B Instruct on context, maximum output, input modes, tool use and reasoning, scores at least as high, and costs no more.

#	Model	Intelligence	$/M	Holds under
1	DeepSeek V4 Flash 0423 DeepSeek	1431.6 +321.9	$0.101	Balanced −16% Chat −42% Code gen −43% Agentic −41%
2	Solar Pro 4 Upstage	1376.2 +266.5	$0.052	Balanced −56% Summarise −57% Chat −64% Code gen −62% Agentic −68%
3	gpt-oss-120b OpenAI	1365.6 +255.9	$0.070	Balanced −41% Summarise −32% Chat −44% Code gen −46% Agentic −38%
4	Granite 4.1 8B IBM	1291.6 +181.9	$0.063	Balanced −48% Summarise −18% Chat −57% Code gen −63% Agentic −37%
5	gpt-oss-20b OpenAI	1287.8 +178.1	$0.055	Balanced −54% Summarise −45% Chat −57% Code gen −59% Agentic −51%
6	Llama 3.1 8B Instruct Meta	1186.7 +77	$0.058	Balanced −52% Summarise −31% Chat −66% Code gen −70% Agentic −57%
7	Hy3 Tencent	1441.2 +331.5	$0.083 agentic	Agentic −10%
8	MiMo-V2.5 Xiaomi	1427.3 +317.6	$0.155 chat	Chat −4% Code gen −2% Agentic −14%
9	GPT-5 Nano OpenAI	1320.3 +210.6	$0.055 summarise	Summarise −15% Agentic −18%

Cheaper and higher-scoring, but you give something up 8

These score at least as high and cost no more on the two plotted axes, and lose something that is not on them. Read the last column before switching.

Model	Intelligence	$/M	Holds under	What you give up
Qwen3.5-Flash Qwen	1397.6 +287.9	$0.114	Balanced −5% Chat −12% Code gen −17%	118K → 66K max output
Qwen3 30B A3B Instruct 2507 Qwen	1384.3 +274.6	$0.084	Balanced −30% Summarise −13% Chat −35% Code gen −38% Agentic −24%	118K → 32K max output
Gemma 3 12B Google	1334.2 +224.5	$0.075	Balanced −38% Summarise −14% Chat −44% Code gen −50% Agentic −29%	118K → 16K max output
Gemma 3 4B Google	1290.8 +181.1	$0.063	Balanced −48% Summarise −18% Chat −57% Code gen −63% Agentic −37%	118K → 16K max output
Mistral Small 3 Mistral	1233.6 +123.9	$0.058	Balanced −52% Summarise −20% Chat −62% Code gen −69% Agentic −41%	131K → 33K context 118K → 16K max output
Phi 4 Microsoft	1216.8 +107.1	$0.087	Balanced −27% Chat −40% Code gen −49% Agentic −12%	131K → 16K context 118K → 15K max output
GLM 4.7 Flash Z.ai	1352.9 +243.2	$0.063 summarise	Summarise −2% Agentic −12%	118K → 16K max output
Qwen3 32B Qwen	1340.1 +230.4	$0.160 chat	Chat −1% Code gen −8%	118K → 16K max output

What this compares, and what it leaves out

Quality is LMArena Elo, used under CC BY 4.0 from the official dataset. The 95% confidence interval on a difference between two scores is about ±10.68 points, so a gap smaller than that is marked tie rather than an improvement — see significance bands.
188 further models at this delivery mode carry a price but no independent score. They are absent from the comparison in both directions — unrated is not a zero, and an unmeasured model is neither an alternative nor a worse buy.
Retired models are never offered as an alternative, and a model only competes against its own delivery mode: batch trades latency for price, so it is not a like-for-like swap.
Nothing here measures latency, throughput, rate limits or how a model behaves on your prompts. Two models with the same index score are not interchangeable.
The models that nothing beats on both axes are on the value frontier, and every other model something cheaper beats is listed here.