Does cheaper mean worse at code?

No. Several budget open-source models (Qwen Coder, DeepSeek Coder, Codestral) match or beat premium models on specific coding benchmarks like HumanEval and MBPP. Premium models like GPT-5 and Claude Opus still win on long-horizon agentic coding and SWE-bench Verified.

Which benchmarks matter for picking a coding model?

SWE-bench Verified measures end-to-end repo patching (the hardest). HumanEval and MBPP measure function-level code completion. LiveCodeBench is the freshest and hardest to game. BigCodeBench covers tool-using code. Pick based on the job.

How do I keep coding costs low on a real project?

Use prompt caching (50% to 90% off cached tokens on Anthropic, OpenAI, DeepInfra). Reserve frontier models (Claude Opus, GPT-5) for hard tasks, route bulk work to cheaper models. See our coding agent stack for the full playbook.

Should I self-host a coding model?

Only if you have GPU capacity sitting idle. For most teams, API access to DeepSeek, Qwen, or Codestral on DeepInfra or Together is cheaper than running your own nodes once you factor in 24/7 cost, spares, and ops.

Use case · Coding

Cheapest coding LLMs

The cheapest models that hold up on SWE-bench, HumanEval, MBPP, LiveCodeBench. Ranked by price per 1M input tokens.

Models30

Cheapest$0.00

UpdatedDaily

All pricing Back to pricing

What this page is

This page ranks every model with a credible coding benchmark score by the cheapest input token price. We include models with scores on SWE-bench, HumanEval, MBPP, LiveCodeBench, BigCodeBench. Open-source coding models (Qwen Coder, DeepSeek Coder, Codestral) dominate the cheap end. Premium reasoning models (Claude Opus, GPT-5) sit at the top of the price scale but win long-horizon agent tasks. Use this list to size cost before locking in a coding stack.

Ranked by input price

Models with meaningful coding benchmark scores, cheapest first.

#	Model	Provider	In $/1M	Out $/1M	Context	avg score	Type
1	gpt-oss-120b (free)	OpenAI	$0.00	$0.00	131K	68.7	OSS
2	gpt-oss-20b (free)	OpenAI	$0.00	$0.00	131K	66.4	OSS
3	gpt-oss-120b	OpenAI	$0.04	$0.18	131K	46.9	OSS
4	Qwen3 30B A3B Instruct 2507	Alibaba Qwen	$0.05	$0.19	131K	55.3	OSS
5	GPT-5 Nano	OpenAI	$0.05	$0.40	400K	43.3	Closed
6	ERNIE 4.5 21B A3B Thinking	baidu	$0.07	$0.28	131K	57.0	OSS
7	Gemma 3 27B	Google DeepMind	$0.08	$0.16	131K	42.2	OSS
8	Qwen3 32B	Alibaba Qwen	$0.08	$0.28	131K	58.2	OSS
9	MiMo-V2-Flash	xiaomi	$0.09	$0.29	262K	73.3	OSS
10	Qwen3 235B A22B Instruct 2507	Alibaba Qwen	$0.09	$0.10	262K	48.5	OSS
11	Qwen3 Next 80B A3B Instruct	Alibaba Qwen	$0.09	$1.10	262K	54.4	OSS
12	Qwen3 Next 80B A3B Thinking	Alibaba Qwen	$0.10	$0.78	262K	61.6	OSS
13	Llama 4 Scout	Meta	$0.10	$0.30	10.0M	18.9	OSS
14	Step 3.5 Flash	stepfun	$0.10	$0.30	262K	76.9	OSS
15	Qwen3 8B	Alibaba Qwen	$0.12	$0.46	131K	56.5	OSS
16	Qwen3 30B A3B Thinking 2507	Alibaba Qwen	$0.13	$1.56	131K	66.2	OSS
17	Hunyuan A13B Instruct	tencent	$0.14	$0.57	131K	52.3	OSS
18	Qwen3 235B A22B Thinking 2507	Alibaba Qwen	$0.15	$1.50	262K	55.9	OSS
19	Llama 4 Maverick	Meta	$0.15	$0.60	1.0M	28.0	OSS
20	MiniMax M2.5	minimax	$0.15	$0.90	205K	55.1	OSS
21	LongCat Flash Chat	meituan	$0.20	$0.80	131K	59.8	OSS
22	DeepSeek V3.2	DeepSeek	$0.21	$0.32	131K	51.2	OSS
23	GPT-5 Mini	OpenAI	$0.25	$2.00	400K	54.3	Closed
24	MiniMax M2	minimax	$0.26	$1.02	205K	69.5	OSS
25	Qwen3.6 Plus	Alibaba Qwen	$0.33	$1.95	1.0M	59.9	OSS
26	Kimi K2.5	moonshotai	$0.38	$2.02	262K	52.0	OSS
27	Qwen3.5 397B A17B	Alibaba Qwen	$0.39	$2.45	256K	78.4	OSS
28	DeepSeek V3.2 Speciale	DeepSeek	$0.40	$1.20	164K	78.2	OSS
29	GLM 4.7	z-ai	$0.40	$1.75	203K	50.5	OSS
30	GPT-4.1 Mini	OpenAI	$0.40	$1.60	1.0M	44.5	Closed

Top 3 cheapest coding LLMs

Best value coding model

gpt-oss-120b (free) is the cheapest model that clears our coding benchmark filter. At $0.00 per 1M input tokens, it lands well below frontier models while keeping credible SWE-bench and HumanEval scores.

gpt-oss-20b (free) follows close behind on price while holding up on coding-specific evals. Good fallback if gpt-oss-120b (free) is down or rate limited.

gpt-oss-120b follows close behind on price while holding up on coding-specific evals. Good fallback if gpt-oss-120b (free) is down or rate limited.

The price gap · cheapest vs most expensive

Cheapest

gpt-oss-120b (free)

$0.00/M

$ per 1M input tokens

Why the gap

DeepSeek V3.2 Speciale is roughly Infinityx the price of gpt-oss-120b (free). The premium pays for long-context agentic coding, higher SWE-bench Verified scores, and vendor reliability. For one-shot code gen or refactors, the cheap options deliver 80 to 95 percent of the quality.

Most expensive

DeepSeek V3.2 Speciale

$0.40/M

$ per 1M input tokens

Frequently asked questions

gpt-oss-120b (free) at $0.00 per 1M input tokens is the cheapest model with a meaningful coding benchmark score. Check the full table above for 30 options ranked by price.

Cheapest coding LLMs

Ranked by input price

Top 3 cheapest coding LLMs

The price gap · cheapest vs most expensive

Frequently asked questions

See also

Related pricing

Stacks

Compare