Cost per task: agentic coding session

One long tool-using loop: 60,000 tokens in across turns, 4,000 tokens out. Cost = input tokens × input price + output tokens × output price. “With caching” assumes three-quarters of the input is served from the prompt cache at the provider's cache-read price.

As of 4 Oct 2026, BenchLeader ranks Gemma 4 E4B first on the Cost per task: agentic coding session board at $0.0016, ahead of Sarvam 30B ($0.0022) and Granite 4.2 3B ($0.0023).

278 models · data as of 4 Oct 2026

1. 1Gemma 4 E4B (thinking)$0.0016
2. 2Sarvam 30B (high)$0.0022
3. 3Granite 4.2 3B$0.0023
4. 4LFM2 24B A2B$0.0023
5. 5Qwen3.5-4B (thinking)$0.0024
6. 6Llama 3.2 1B Instruct$0.0024
7. 7Gemma 4 E2B (thinking)$0.0027
8. 8Nova Micro$0.0027
9. 9Granite 3.3 8B (none)$0.0028
10. 10Mercury 2.5$0.0030
11. 11Sarvam 105B (high)$0.0031
12. 12Llama 3.1 8B$0.0032
13. 13Mistral Small 3$0.0033
14. 14Gemma 3 4B$0.0034
15. 15Gemma 3 12B$0.0036

SearchAll providersAi2AI21 LabsAlibabaAmazonAnt GroupAnthropicApodexArcee AIBaiduByteDanceCelerisCohereDeep CogitoDeepSeekGoogleIBMInceptionKwaiKATLiquid AIMeituanMetaMicrosoftMiniMaxMistral AIMoonshot AIMultiverse ComputingNex AGINous ResearchNVIDIAOpenAIRekaSarvamSpaceXAIStepFunSwiss AIThinking MachinesUpstageXiaomiZhipu AIOpen weightsNew models278 of 278

#	Model	$ per task ↓	With caching ↕	Index ↕
1	GPT-5.5 ProxhighOpenAI	$2.52	–	65.6
2	GPT-5.4 ProOpenAI	$2.52	–	64.0
3	GPT-4OpenAI	$2.04	–	44.3
4	GPT 4 0613OpenAI	$2.04	–	39.9
5	o3-prohighOpenAI	$1.52	–	54.3
6	GPT-5 ProOpenAI	$1.38	–	60.2
7	Claude Opus 4.1thinkingAnthropic	$1.20	$0.593	56.6
8	Claude Opus 4thinkingAnthropic	$1.20	$0.593	55.3
9	Claude 3 OpusAnthropic	$1.20	–	39.3
10	o1OpenAI	$1.14	$0.802	53.3
11	GPT-6 AstramaxOpenAI	$0.800	$0.395	70.2
12	Claude Fable 5.1xhighAnthropic	$0.800	$0.361	70.0
13	Claude Fable 5thinkingAnthropic	$0.800	$0.395	69.2
14	GPT-4 TurboOpenAI	$0.720	–	41.2
15	GPT-5.5xhighOpenAI	$0.420	$0.217	66.2
16	GPT-5.5 InstantOpenAI	$0.420	–	58.1
17	Claude Opus 5highAnthropic	$0.400	$0.198	68.4
18	Claude Opus 4.7Anthropic	$0.400	$0.198	64.0
19	Claude Opus 4.6Anthropic	$0.400	$0.198	63.3
20	Claude Opus 4.8maxAnthropic	$0.400	$0.198	62.5
21	Claude Opus 4.5thinkingAnthropic	$0.400	$0.198	60.2
22	Claude Opus 5.5thinkingAnthropic	$0.320	$0.149	70.1
23	GPT-5.6 SolmaxOpenAI	$0.320	$0.158	67.5
24	Grok 3SpaceXAI	$0.320	–	51.6
25	Kimi K3maxMoonshot AIopen ↗	$0.240	$0.118	65.6
26	Claude Sonnet 4.6Anthropic	$0.240	$0.118	58.7
27	Claude Sonnet 4.5highAnthropic	$0.240	$0.118	58.4
28	Grok 4SpaceXAI	$0.240	–	58.1
29	Claude Sonnet 4thinkingAnthropic	$0.240	$0.118	53.5
30	Claude 3.5 SonnetAnthropic	$0.240	–	46.4
31	GPT-5.4xhighOpenAI	$0.210	$0.109	64.4
32	Nova PremierAmazon	$0.200	$0.116	45.4
33	GPT-4oOpenAI	$0.190	$0.134	44.1
34	Command ACohereopen ↗	$0.190	–	42.0
35	Qwen3 7maxAlibaba	$0.180	$0.090	60.0
36	GPT-5.6 TerramaxOpenAI	$0.168	$0.087	63.7
37	Gemini 3.1 ProGoogle	$0.168	$0.087	62.4
38	GPT-5.3-CodexxhighOpenAI	$0.161	$0.090	64.0
39	GPT-5.2xhighOpenAI	$0.161	$0.090	60.9
40	GPT-5.2-CodexxhighOpenAI	$0.161	$0.090	59.6
41	NewGPT-6.1 SolmaxOpenAI	$0.160	$0.074	69.5
42	NewGemini 4 ArgonhighGoogle	$0.160	–	69.5
43	NewClaude Sonnet 5.5xhighAnthropic	$0.160	$0.079	67.0
44	GPT-6 SolmaxOpenAI	$0.160	$0.079	66.3
45	Claude Sonnet 5highAnthropic	$0.160	$0.079	60.4
46	o3OpenAI	$0.152	$0.085	59.7
47	GPT-4.1OpenAI	$0.152	$0.085	48.6
48	Jamba 1.5 LargeAI21 Labsopen	$0.152	–	42.4
49	Jamba Large 1.6AI21 Labs	$0.152	–	26.0
50	Qwen3.8 MaxmaxAlibaba	$0.144	$0.065	65.0
51	Grok 4.6mediumSpaceXAI	$0.144	$0.076	64.4
52	Qwen3.8 2.4T A95BAlibabaopen ↗	$0.144	$0.065	63.5
53	Grok 4.7lowSpaceXAI	$0.144	$0.076	61.7
54	Grok 4.5highSpaceXAI	$0.144	$0.068	59.6
55	Mimo v2 ProXiaomi	$0.144	$0.072	58.9
56	Pixtral LargeMistral AIopen	$0.144	–	42.0
57	Mistral Large 2Mistral AIopen	$0.144	$0.056	40.7
58	Mixtral 8x22B InstructMistral AIopen ↗	$0.144	–	37.6
59	Gemini 3.5 FlashmediumGoogle	$0.126	$0.065	63.1
60	Mistral Medium 3.5Mistral AIopen	$0.120	$0.059	49.6

Show 100 more

Cite as: BenchLeader, “Cost per task: agentic coding session”, https://www.benchleader.com/leaderboards/cost-agent-session, data as of 4 Oct 2026.