Cheaper alternatives to Claude 3 Haiku

Anthropic LMArena 1194.8 $0.500/M on a balanced workload prices as of 2026-08-28

11 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, 48 score at least as high as Claude 3 Haiku and cost no more under at least one workload. 47 of them are genuinely cheaper; the rest match the price and win on score alone. 11 match or beat it on every capability we hold — context, maximum output, input modes, tool use and extended reasoning. 37 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 Claude 3 Haiku, it is not generated.

Where it holds

Workload	Claude 3 Haiku $/M	Better and cheaper	With a trade
Balanced	$0.500	10	33
Summarise	$0.237	9	24
Chat	$0.584	9	34
Code gen	$0.832	10	37
Agentic	$0.269	9	26

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	1194.8
Context window	200K
Max output	4K
Input modes	text, image
Tool use	yes
Extended reasoning	no

A replacement has to clear every line above, not just the score. Full record for Claude 3 Haiku.

Better and cheaper, nothing given up 11

Each of these matches or beats Claude 3 Haiku 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	Gemma 4 31B Google	1441.7 +246.9	$0.152	Balanced −70% Summarise −62% Chat −70% Code gen −72% Agentic −61%
2	Gemma 4 26B A4B Google	1434.6 +239.8	$0.138	Balanced −73% Summarise −65% Chat −70% Code gen −72% Agentic −59%
3	MiMo-V2.5 Xiaomi	1427.3 +232.5	$0.175	Balanced −65% Summarise −55% Chat −73% Code gen −74% Agentic −70%
4	GPT-5.6 Luna OpenAI	1428.5 +233.7	$0.450	Balanced −10% Summarise −16% Chat −7% Code gen −6% Agentic −10%
5	Qwen3.5-Flash Qwen	1397.6 +202.8	$0.114	Balanced −77% Summarise −69% Chat −76% Code gen −78% Agentic −65%
6	Qwen3.5-35B-A3B Qwen	1395.6 +200.8	$0.500	Balanced same price
7	GPT-5.4 Nano OpenAI	1372.8 +178	$0.463	Balanced −7% Summarise −15% Chat −3% Code gen −2% Agentic −7%
8	Gemma 3 27B Google	1358.3 +163.5	$0.172	Balanced −66% Summarise −63% Chat −63% Code gen −64% Agentic −58%
9	GPT-5 Nano OpenAI	1320.3 +125.5	$0.138	Balanced −73% Summarise −77% Chat −70% Code gen −69% Agentic −72%
10	GPT-4.1 Nano OpenAI	1284.8 +90	$0.175	Balanced −65% Summarise −61% Chat −66% Code gen −67% Agentic −63%
11	MiniMax M3 MiniMax	1434.8 +240	$0.821 code gen	Code gen −1%

Cheaper and higher-scoring, but you give something up 37

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
Hy3 Tencent	1441.2 +246.4	$0.144	Balanced −71% Summarise −67% Chat −72% Code gen −73% Agentic −69%	no image input
DeepSeek V4 Flash 0423 DeepSeek	1431.6 +236.8	$0.101	Balanced −80% Summarise −72% Chat −84% Code gen −85% Agentic −80%	no image input
DeepSeek V3.2 DeepSeek	1424.6 +229.8	$0.290	Balanced −42% Summarise −4% Chat −54% Code gen −61% Agentic −25%	200K → 164K context no image input
DeepSeek V3.2 Exp DeepSeek	1424.4 +229.6	$0.305	Balanced −39% Chat −44% Code gen −57%	200K → 164K context no image input
Qwen3 235B A22B Instruct 2507 Qwen	1419.3 +224.5	$0.205	Balanced −59% Summarise −52% Chat −53% Code gen −56% Agentic −41%	no image input
Qwen3 Next 80B A3B Instruct Qwen	1418.6 +223.8	$0.350	Balanced −30% Summarise −40% Chat −16% Code gen −16% Agentic −14%	no image input
DeepSeek V3.1 Terminus DeepSeek	1419.6 +224.8	$0.453	Balanced −9% Chat −11% Code gen −16%	200K → 164K context no image input
Step 3.5 Flash StepFun	1403.8 +209	$0.150	Balanced −70% Summarise −54% Chat −69% Code gen −74% Agentic −52%	no image input
Qwen3 30B A3B Instruct 2507 Qwen	1384.3 +189.5	$0.084	Balanced −83% Summarise −77% Chat −82% Code gen −84% Agentic −74%	no image input
GLM 4.5 Air Z.ai	1382.8 +188	$0.310	Balanced −38% Summarise −43% Chat −34% Code gen −33% Agentic −35%	200K → 131K context no image input
Solar Pro 4 Upstage	1376.2 +181.4	$0.052	Balanced −90% Summarise −88% Chat −90% Code gen −90% Agentic −89%	no image input
DeepSeek V3 0324 DeepSeek	1375 +180.2	$0.438	Balanced −13% Chat −6% Code gen −16%	200K → 164K context no image input
GLM 4.6V Z.ai	1374.7 +179.9	$0.450	Balanced −10% Chat −20% Code gen −23% Agentic −9%	200K → 131K context
gpt-oss-120b OpenAI	1365.6 +170.8	$0.070	Balanced −86% Summarise −82% Chat −85% Code gen −86% Agentic −79%	200K → 131K context no image input
Qwen3 Next 80B A3B Thinking Qwen	1367.5 +172.7	$0.412	Balanced −18% Summarise −15% Chat −2% Code gen −6%	no image input
Mercury 2 Inception	1357.8 +163	$0.375	Balanced −25% Summarise −11% Chat −35% Code gen −36% Agentic −29%	200K → 128K context no image input
GLM 4.7 Flash Z.ai	1352.9 +158.1	$0.145	Balanced −71% Summarise −74% Chat −69% Code gen −69% Agentic −70%	no image input
MiniMax M2.5 MiniMax	1359 +164.2	$0.472	Balanced −6% Chat −11% Code gen −12% Agentic −8%	no image input
Qwen3 32B Qwen	1340.1 +145.3	$0.130	Balanced −74% Summarise −62% Chat −73% Code gen −76% Agentic −59%	200K → 131K context no image input
Trinity Large Thinking Arcee	1341.9 +147.1	$0.378	Balanced −25% Summarise −13% Chat −27% Code gen −30% Agentic −19%	no image input
MiniMax M2 MiniMax	1342.1 +147.3	$0.446	Balanced −11% Chat −4% Code gen −14%	no image input
Gemma 3 12B Google	1334.2 +139.4	$0.075	Balanced −85% Summarise −77% Chat −85% Code gen −87% Agentic −76%	200K → 131K context
DeepSeek V3 DeepSeek	1332.6 +137.8	$0.450	Balanced −10% Chat −3% Code gen −13%	200K → 164K context no image input
Qwen3 30B A3B Qwen	1316.9 +122.1	$0.215	Balanced −57% Summarise −41% Chat −53% Code gen −58% Agentic −34%	200K → 131K context no image input
Olmo 3 32B Think Allen AI	1298.5 +103.7	$0.237	Balanced −53% Summarise −29% Chat −50% Code gen −57% Agentic −25%	200K → 66K context no image input no tool use
Granite 4.1 8B IBM	1291.6 +96.8	$0.063	Balanced −88% Summarise −78% Chat −88% Code gen −90% Agentic −79%	200K → 131K context no image input
Gemma 3 4B Google	1290.8 +96	$0.063	Balanced −88% Summarise −78% Chat −88% Code gen −90% Agentic −79%	200K → 131K context no tool use
gpt-oss-20b OpenAI	1287.8 +93	$0.055	Balanced −89% Summarise −85% Chat −88% Code gen −89% Agentic −83%	200K → 131K context no image input
GPT-4o-mini (2024-07-18) OpenAI	1286.6 +91.8	$0.263	Balanced −48% Summarise −36% Chat −47% Code gen −50% Agentic −36%	200K → 128K context
Qwen2.5 72B Instruct Qwen	1269.1 +74.3	$0.370	Balanced −26% Chat −36% Code gen −54%	200K → 33K context no image input
Llama 3.1 70B Instruct Meta	1261 +66.2	$0.400	Balanced −20% Chat −32% Code gen −52%	200K → 131K context no image input
Mistral Small 3 Mistral	1233.6 +38.8	$0.058	Balanced −89% Summarise −78% Chat −89% Code gen −92% Agentic −80%	200K → 33K context no image input no tool use
Phi 4 Microsoft	1216.8 +22	$0.087	Balanced −83% Summarise −69% Chat −83% Code gen −87% Agentic −70%	200K → 16K context no image input no tool use
MiMo-V2.5-Pro Xiaomi	1465 +270.2	$0.480 chat	Chat −18% Code gen −21% Agentic −9%	no image input
MiniMax M2.7 MiniMax	1405.3 +210.5	$0.821 code gen	Code gen −1%	no image input
Llama 3.3 70B Instruct Meta	1274.9 +80.1	$0.710 code gen	Code gen −15%	200K → 131K context no image input
Gemma 2 27B Google	1231.5 +36.7	$0.650 code gen	Code gen −22%	200K → 8K context 4K → 2K max output no image input no tool use

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.