TerminalBench Hard benchmark
309 AI models ranked on TerminalBench Hard. Each row shows the model’s score next to its cheapest API price per 1M tokens where it has one, so you can weigh quality against cost.
Top model today
GPT-5.6 Sol (OpenAI) leads at 66%, from $2.00 per 1M input OpenRouter routed quote.
Data via GitHub: laude-institute/terminal-bench →Models ranked
309
on this benchmark
Top score
66%
current leader
Type
Single eval
benchmark
Updated
Oct 14, 2026
benchmark scores
Cost vs TerminalBench Hard
198 priced models = reasoning model (extended thinking)
Each labelled name links to that model’s page. Curated to the cost-efficiency frontier, top scorers and cheapest so labels stay readable; the full field is in the leaderboard table below (and via “select all”). Reasoning models (extended “thinking” before answering) carry a ring around the dot.
Models ranked on TerminalBench Hard
highest score first · 309 modelsMaker
| # | ||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | GPT-5.6 Sol |  OpenAI | 66% | $2.00 Routed quote | $10.00 Routed quote | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 2 | Claude Fable 5 |  Anthropic | 63% | $10.00 via Anthropic | $50.00 via Anthropic | |||||||||||||||||||||||||
| 3 | GPT-5.6 Terra | OpenAI | 63%xhigh | $2.00 via OpenAI | $12.00 via OpenAI | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 4 | GPT-5.5 | OpenAI | 61% | $5.00 via OpenAI | $30.00 via OpenAI | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 5 | Claude Opus 4.8 | Anthropic | 58% | $5.00 via Anthropic | $25.00 via Anthropic | |||||||||||||||||||||||||
| 6 | GPT-5.4 | OpenAI | 58% | $2.50 via Databricks | $15.00 via OpenAI | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 7 | Claude Opus 4.7 | Anthropic | 55%reasoning: false | $5.00 via Anthropic | $25.00 via Anthropic | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 8 | Gemini 3.1 Pro Preview |  Google | 54% | $2.00 via Google | $12.00 via Google | |||||||||||||||||||||||||
| 9 | Claude Sonnet 4.6 | Anthropic | 53%reasoning: adaptive | $3.00 via Databricks | $15.00 via Anthropic | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 10 | GPT-5.3-Codex | OpenAI | 53% | $1.75 via Aihubmix | $14.00 via Aihubmix | |||||||||||||||||||||||||
| 11 | GPT-5.4 Mini | OpenAI | 52% | $0.75 via Databricks | $4.50 via OpenAI | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 12 | GLM 5.2 |  Zhipu | 51% | $0.05 Routed quote | $1.98 via Scx AI | |||||||||||||||||||||||||
| 13 | Qwen3.7 Max |  Alibaba | 51% | $1.25 via Novita | $3.75 via Novita | |||||||||||||||||||||||||
| 14 | KAT-Coder-Pro V2 | Other | 49% | $0.30 Routed quote | $1.20 Routed quote | |||||||||||||||||||||||||
| 15 | Claude Opus 4.6 | Anthropic | 48% | $5.00 via Anthropic | $25.00 via Anthropic | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 16 | Claude Opus 4.5 | Anthropic | 47%reasoning: true | $5.00 via Anthropic | $25.00 via Anthropic | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 17 | GPT-5.2 | OpenAI | 47% | $1.75 via OpenAI | $14.00 via OpenAI | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 18 | Qwen3.7 Plus | Alibaba | 47% | $0.282 via Aihubmix | $1.13 via Aihubmix | |||||||||||||||||||||||||
| 19 | DeepSeek V4 Pro 0423 |  DeepSeek | 46% | $0.435 via Tencent | $0.87 via Tencent | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 20 | Gemini 3.5 Flash | Google | 46%minimal | $1.50 via Google | $9.00 via Google | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 21 | GPT-5.1 | OpenAI | 45% | $1.25 via Databricks | $10.00 via Databricks | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 22 | Muse Spark |  Meta | 45% | n/a | n/a | |||||||||||||||||||||||||
| 23 | Kimi K2.7 Code |  Moonshot | 45% | $0.671 Routed quote | $3.35 Routed quote | |||||||||||||||||||||||||
| 24 | Kimi K2.6 | Moonshot | 44% | $0.65 via Wandb | $3.40 via Novita | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 25 | Qwen3.6 Plus | Alibaba | 44% | $0.325 Routed quote | $1.95 Routed quote | |||||||||||||||||||||||||
| 26 | Qwen3.6 Max Preview | Alibaba | 44% | n/a | n/a | |||||||||||||||||||||||||
| 27 | GLM 5.1 | Zhipu | 43% | $1.05 via DeepInfra | $3.50 via DeepInfra | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 28 | GLM 5 | Zhipu | 43% | $0.60 via DeepInfra | $1.92 Routed quote | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 29 | MiMo-V2.5-Pro |  Xiaomi | 43% | $0.435 Routed quote | $0.87 Routed quote | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 30 | GPT 5.4 Low | OpenAI | 43% | $2.50 via Aihubmix | $15.00 via Aihubmix | |||||||||||||||||||||||||
| 31 | MiniMax M3 |  MiniMax | 42% | $0.23 via Wandb | $0.96 via Wandb | |||||||||||||||||||||||||
| 32 | GPT-5.4 Nano | OpenAI | 42% | $0.20 via Databricks | $1.25 via Databricks | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 33 | GPT-5.5 Instant (May 2026) | OpenAI | 42% | n/a | n/a | |||||||||||||||||||||||||
| 34 | MiMo-V2.5 | Xiaomi | 42% | $0.14 Routed quote | $0.28 Routed quote | |||||||||||||||||||||||||
| 35 | Gemini 3 Pro | Google | 42% | $2.00 via Replicate | $12.00 via Replicate | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 36 | Qwen3.5 397B A17B | Alibaba | 41% | $0.164 via Aihubmix | $0.986 via Aihubmix | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 37 | Mimo V2 Pro | Xiaomi | 41% | $1.10 via Aihubmix | $3.30 via Aihubmix | |||||||||||||||||||||||||
| 38 | Grok 4 20 0309 |  xAI | 41% | $1.25 via xAI | $2.50 via xAI | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 39 | MiniMax M2.7 | MiniMax | 39% | $0.25 via DeepInfra | $0.55 via Pinstripes | |||||||||||||||||||||||||
| 40 | DeepSeek V4 Flash 0423 | DeepSeek | 39%high | $0.09 via DeepInfra | $0.18 via DeepInfra | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 41 | Gemini 3 Flash | Google | 39%reasoning: true | $0.50 Routed quote | $3.00 Routed quote | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 42 | GPT-5 | OpenAI | 38%medium | $1.25 via Databricks | $10.00 via Databricks | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 43 | Grok 4 | xAI | 38% | $1.25 via xAI | $2.50 via xAI | |||||||||||||||||||||||||
| 44 | Grok 4.3 | xAI | 38% | $1.25 via xAI | $2.50 via xAI | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
| 45 | GPT 5 Codex | OpenAI | 38% | $1.25 Routed quote | $10.00 Routed quote | |||||||||||||||||||||||||
| 46 | Grok 4.20 | xAI | 38% | $1.25 via xAI | $2.50 via xAI | |||||||||||||||||||||||||
| 47 | GPT-5.2-Codex | OpenAI | 37% | $1.75 via Databricks | $14.00 via Databricks | |||||||||||||||||||||||||
| 48 | o3 | OpenAI | 37% | $2.00 via OpenAI | $8.00 via OpenAI | |||||||||||||||||||||||||
| 49 | Nvidia Nemotron 3 Ultra 550B A55b |  NVIDIA | 36% | $0.50 via DeepInfra | $2.15 via Wandb | |||||||||||||||||||||||||
| 50 | Gemma 4 31B | Google | 36% | $0.14 via Amazon Bedrock | $0.40 via Amazon Bedrock | |||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||
1–50 of 310
Page 1 of 7
On-demand listed rates, OpenRouter routed quotes and Vercel reseller quotes. Enterprise commitments can be materially lower. Cost per task uses best-effort token usage. An OpenRouter routed quote is the price of whichever upstream provider OpenRouter prefers today, so it moves day to day.
Frequently asked questions
What is the TerminalBench Hard benchmark?
The hardest subset of the Terminal-Bench agentic benchmark, made up of the tasks frontier models complete least often across software engineering, sysadmin, data processing, ML and security. An agent is given an English task and a sandboxed container terminal, and is scored by the pass rate of automated verification tests.
Which AI model scores highest on TerminalBench Hard?
GPT-5.6 Sol (OpenAI) leads with 66%, from $2.00 per 1M input tokens as an OpenRouter routed quote, while OpenAI lists it at $4.00 per 1M input.
How many models are ranked on TerminalBench Hard?
309 models carry a TerminalBench Hard score. Scores come from Artificial Analysis; where a model has a listed API price it is drawn from LiteLLM and OpenRouter, and models with no current price are listed without one.
Other benchmarks
compare the same models on a different evalIntelligence Indexcomposite indexCoding Indexcomposite indexAgentic Indexcomposite indexGPQA DiamondleaderboardHumanity's Last ExamleaderboardMMLU-ProleaderboardMMMU-ProleaderboardLiveCodeBenchleaderboardSciCodeleaderboardTerminalBench v2.1leaderboardτ²-benchleaderboardτ-bench BankingleaderboardAPEX AgentsleaderboardIT-Bench SREleaderboardGDPvalleaderboardCritPtleaderboardOmniscienceleaderboardIFBenchleaderboardLong-Context Reasoningcomposite indexSWE-bench VerifiedleaderboardSWE-bench ProleaderboardAIME 2026leaderboardARC-AGI-2leaderboardAA-Briefcaseleaderboard
Benchmark scores via Artificial Analysis · scores updated Oct 14, 2026 · Listed prices from LiteLLM, OpenRouter and Vercel · catalogue refreshed Oct 14, 2026 · 123 prices older, back to Jul 19, 2026
Every weekday
AI moves fast. Here's your debrief.
News, analysis, tools, and more.
For people who build with AI