Which AI model is best at agentic coding in October 2026?

Claude Sonnet 5.5 passes the most Terminal-Bench 4.0 tasks in Artificial Analysis's runs, 63.6%, followed by Claude Opus 5.5 (59.6%) and GPT-6 Astra (59.1%). On our measured coding-agent traffic Claude Sonnet 5.5 costs $0.053 per request.

Source: Artificial Analysis (artificialanalysis.ai). Fetched October 13, 2026. Cost per request: Standard Compute gateway data, 79,301 coding-agent requests, Sep 13 to Oct 12, 2026. How we compute this

Run current Claude, GPT, Gemini and open models in your agent from $19/month

See plans

27 current models ranked by Terminal-Bench 4.0

Model	Terminal-Bench 4.0Artificial Analysis	Intelligence IndexArtificial Analysis	Cost per requestour data
1. Claude Sonnet 5.5max effort	63.6%	56	$0.053
2. Claude Opus 5.5max effort	59.6%	58	$0.11
3. GPT-6 Astramax effort	59.1%	53	$0.41
4. Gemini 4 Argonhigh effort	57.1%	53	$0.11
5. GPT-6.1 Solmax effort	56.1%	52	$0.067
6. Muse Wondermax effort	52.0%	52	$0.11
6. Claude Fable 5.1max effort	52.0%	53	$0.24
8. GLM 5.3max effort	41.9%	45	$0.057
9. GPT-5.6 Terramax effort	35.4%	42	$0.084
10. Step 5 Preview	33.3%	44	$0.025
11. Claude Haiku 5.5max effort	32.8%	43	$0.013
12. DeepSeek V4.1 Flashmax effort	26.8%	39	$0.0067
12. Mistral Large 4Mistral Large 4 Preview	26.8%	38	$0.042
14. Grok 4.7xhigh effort	25.8%	46	$0.13
15. Gemini 3.8 Flashhigh effort	19.7%	41	$0.048
16. DeepSeek V4 ProDeepSeek V4 Pro 0813, max effort	14.1%	36	$0.030
17. GPT-6 Lunamax effort	12.6%	38	$0.0041
18. Gemini 3.1 ProGemini 3.1 Pro Preview	4.0%	30	$0.095
19. MiniMax M3	2.0%	29	$0.013
19. GPT-5.4 Minixhigh effort	2.0%	24	$0.023
21. Qwen3.7 Max	1.5%	29	$0.063
22. Qwen3.7 Plus	1.0%	25	$0.014
22. Kimi K2.7 Code	1.0%	26	$0.027
24. Gemini 3.1 Flash Lite	0.5%	16	$0.0082
24. Kimi K2.6reasoning	0.5%	27	$0.024
26. Llama 4 Maverick	0.0%	10	$0.0095
26. Devstral 2	0.0%	9	$0.010

Source: Artificial Analysis (artificialanalysis.ai). Each model is measured as the variant named under it. Cost per request is ours: every one of 79,301 real coding-agent requests priced at list rates (OpenRouter's listed price for some older and open models). Methodology

Older versions

Model	Terminal-Bench 4.0Artificial Analysis	Intelligence IndexArtificial Analysis	Cost per requestour data
GPT-6 Solsuperseded by GPT-6.1 Sol	43.9%	48	$0.083
Claude Fable 5superseded by Claude Fable 5.1	42.4%	50	$0.32
GPT-5.6 Solsuperseded by GPT-6.1 Sol	39.9%	47	$0.17
Claude Opus 4.8superseded by Claude Opus 5.5	21.7%	42	$0.16
GPT-5.5superseded by GPT-6.1 Sol	14.6%	38	$0.21
GPT-5.6 Lunasuperseded by GPT-6 Luna	11.6%	37	$0.0084
Gemini 3.5 Flashsuperseded by Gemini 3.8 Flash	6.6%	33	$0.049
Claude Sonnet 4.6superseded by Claude Sonnet 5.5	3.0%	30	$0.097
GLM 5.2superseded by GLM 5.3	1.0%	34	$0.057
Mistral Large 3superseded by Mistral Large 4	0.0%	9	$0.012
Claude Haiku 4.5superseded by Claude Haiku 5.5	0.0%	17	$0.022
Grok 4.3superseded by Grok 4.7	0.0%	25	$0.068

Source: Artificial Analysis (artificialanalysis.ai). Cost per request is ours. Methodology

FAQ

What is Terminal-Bench 4.0?

A benchmark of agentic coding and terminal tasks from the Laude Institute: the model works in a real shell to build, debug and configure software, and passes a task only when its checks succeed. Artificial Analysis runs it for every model as part of its Intelligence Index.

Why do these scores differ from the tbench.ai leaderboard?

The tbench.ai leaderboard lists agent and model combinations, each run with its own agent harness. Artificial Analysis runs every model the same way, so its percentages are lower for some models and the order can change. Both are useful; compare models within one source.

Which cheap model is best for coding agents?

Among models under $0.02 per coding-agent request on our traffic, Claude Haiku 5.5 scores highest on Terminal-Bench 4.0 in Artificial Analysis's runs (32.8%), at $0.013 per request.

Current Claude, GPT, Gemini and open models are included with Standard Compute

Standard Compute

Current frontier models Claude Opus 5.5 · Claude Sonnet 5.5