korshunov.ai Everything that matters in ML
EN
Benchmark · agentic

Terminal-Bench 2

42 results 33 models

Harbor/Terminus-2 89-task terminal-agent suite; 2.0 and 2.1 are the same task set, so one series. File here only when the quote says 2.0 / 2.1 / v2 — not the original suite (terminal-bench), and not the harder 'Terminal-Bench Hard' subset (not in this catalog).

0 25 50 75 100 2025-11-19 2026-04-15 2026-09-10 GPT-5.1-Codex-Max · 58.1 · 2025-11-19 Step 3.5 Flash · 51 · 2026-02-10 MiniMax M2.7 · 57 · 2026-04-12 GPT-5.5 · 82.7 · 2026-04-23 GPT-5.5 · 82.7 · 2026-04-23 GPT-5.5 · 82.7 · 2026-04-25 Qwen3.6-27B · 59.3 · 2026-04-25 Qwen3.6-27B · 60.7 · 2026-08-10 Tmax · 27 · 2026-06-23 Tmax-27B · 43 · 2026-06-23 GLM 5.2 FP8 with FP8 KV · 79.8 · 2026-07-05 LLM-as-a-Verifier · 86.5 · 2026-07-07 Grok 4.5 · 83.3 · 2026-07-08 hermes · 55 · 2026-07-08 Gemini 3.5 Flash-Lite · 54 · 2026-07-21 Gemini 3.5 Flash-Lite · 54 · 2026-07-21 Gemini 3.5 Flash-Lite · 54 · 2026-07-27 KAT-Coder-V2.5 · 60.7 · 2026-07-26 DeepSeek-V4-Flash · 82.7 · 2026-07-31 DeepSeek-V4-Flash · 79 · 2026-08-01 DeepSeek-V4-Flash-0731 · 82.7 · 2026-07-31 DeepSeek-V4-Flash-0731 · 82.7 · 2026-08-07 V4 Flash 0731 · 79 · 2026-08-01 Inkling-Small · 64.7 · 2026-08-03 Qwen3.8-Max · 86.6 · 2026-08-03 Qwen3.8-Max · 67.4 · 2026-08-05 Muse Spark 1.1 · 80 · 2026-08-05 Pokee-Isaac 28B · 65.1 · 2026-08-08 DeepSeek V4 Flash 0731 · 82.7 · 2026-08-09 Opus 5 · 86.7 · 2026-08-11 Grok 4.6 · 88.4 · 2026-08-13 Grok 4.6 · 88.4 · 2026-08-21 DeepSeek-V4-Pro · 87.9 · 2026-08-13 Ornith-1.5 · 86.1 · 2026-08-19 agents · 94 · 2026-08-20 DeepSeek-V4-Flash-Vision-Exp · 83.9 · 2026-08-21 Granite 4.2 30B · 29.2 · 2026-08-26 Granite 4.2 8B · 20.6 · 2026-08-26 Muse Spark 1.3 · 88.8 · 2026-09-04 K2-Horizon-MoVA-36B-A4B · 58.6 · 2026-09-07 K2-Horizon-375B-A23B · 66.9 · 2026-09-07 DeepSeek-V4.1-Flash · 90.6 · 2026-09-10
GPT-5.1-Codex-Max Step 3.5 Flash MiniMax M2.7 GPT-5.5 Qwen3.6-27B Tmax Tmax-27B GLM 5.2 FP8 with FP8 KV LLM-as-a-Verifier Grok 4.5 hermes Gemini 3.5 Flash-Lite KAT-Coder-V2.5 DeepSeek-V4-Flash DeepSeek-V4-Flash-0731 V4 Flash 0731 Inkling-Small Qwen3.8-Max Muse Spark 1.1 Pokee-Isaac 28B DeepSeek V4 Flash 0731 Opus 5 Grok 4.6 DeepSeek-V4-Pro Ornith-1.5 agents DeepSeek-V4-Flash-Vision-Exp Granite 4.2 30B Granite 4.2 8B Muse Spark 1.3 K2-Horizon-MoVA-36B-A4B K2-Horizon-375B-A23B DeepSeek-V4.1-Flash
Timeline
Date Model Score Source
2026-09-10 DeepSeek-V4.1-Flash 90.6% DeepSeek releases V4.1-Flash model with native multimodal support
2026-09-07 K2-Horizon-MoVA-36B-A4B 58.6% IFM releases K2 Horizon: six Apache 2.0 models from 0.9B to 375B
2026-09-07 K2-Horizon-375B-A23B 66.9% IFM releases K2 Horizon: six Apache 2.0 models from 0.9B to 375B
2026-09-04 Muse Spark 1.3 88.8% Meta releases Muse Spark 1.3 with fewer tool calls and tokens
2026-08-26 Granite 4.2 30B 29.24% IBM releases Granite 4.2 with native reasoning and agentic RL for open enterprise models
2026-08-26 Granite 4.2 8B 20.56% IBM releases Granite 4.2 with native reasoning and agentic RL for open enterprise models
2026-08-21 Grok 4.6 88.4% SpaceXAI introduces Grok 4.6 with Cursor data for agentic work
2026-08-21 DeepSeek-V4-Flash-Vision-Exp 83.9% DeepSeek releases DeepSeek-V4-Flash-Vision-Exp multimodal model
2026-08-20 agents 94.0% Meta releases Muse Video with native audio; Stripe acquires OpenRouter
2026-08-19 Ornith-1.5 86.1% Ornith-1.5 releases 9B, 35B-A3B, and 397B models with self-improving training
2026-08-13 DeepSeek-V4-Pro 87.9% DeepSeek-V4-Pro GA release enhances agent capabilities and adds Responses API support
2026-08-13 Grok 4.6 88.4% xAI releases Grok 4.6; Alibaba opens Qwen3.8-MoE; DeepSeek V4 Pro GA
2026-08-11 Opus 5 86.74% Ouroboros self-developing agent sets new benchmarks with Opus 5
2026-08-10 Qwen3.6-27B 60.7% Meta releases Muse Glimmer, a 30B open-weights agentic model running on one consumer GPU
2026-08-09 DeepSeek V4 Flash 0731 82.7% DeepSeek V4 Flash 0731 matches reported 82.7% on Terminal-Bench 2.1 in public harness
2026-08-08 Pokee-Isaac 28B 65.1% Pokee AI releases Pokee-Isaac 28B, a 10M-token context model for in-boundary deployment
2026-08-07 DeepSeek-V4-Flash-0731 82.7% DeepSeek releases V4-Flash-0731, outperforming V4-Pro at lower cost
2026-08-05 Muse Spark 1.1 80.0% Meta releases Muse Code beta terminal agent powered by Muse Spark 1.2
2026-08-05 Qwen3.8-Max 67.4% Alibaba announces Qwen3.8-Max with 2.4T parameters and upcoming open weights
2026-08-03 Qwen3.8-Max 86.6% Alibaba releases Qwen3.8-Max, a 2.4T-parameter MoE model
2026-08-03 Inkling-Small 64.7% Thinking Machines Lab releases Inkling-Small, a 276B open weights multimodal MoE model
2026-08-01 V4 Flash 0731 79.0% DeepSeek releases V4-Flash with major post-training gains and open weights
2026-08-01 DeepSeek-V4-Flash 79.0% DeepSeek releases V4-Flash with post-training gains and open weights
2026-07-31 DeepSeek-V4-Flash-0731 82.7% DeepSeek releases V4-Flash-0731 with major agentic gains at lower cost
2026-07-31 DeepSeek-V4-Flash 82.7% DeepSeek releases DeepSeek-V4-Flash in public beta with enhanced agent capabilities
2026-07-27 Gemini 3.5 Flash-Lite 54.0% Google releases Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
2026-07-26 KAT-Coder-V2.5 60.7% KwaiKAT releases KAT-Coder-V2.5, an agentic coding model trained on 100,000+ verifiable environments
2026-07-21 Gemini 3.5 Flash-Lite 54.0% Google releases Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber for agentic workloads
2026-07-21 Gemini 3.5 Flash-Lite 54.0% Google releases Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
2026-07-08 hermes 55.0% User benchmarks Mimo v2.5 against DeepSeek V4 Flash and Hermes
2026-07-08 Grok 4.5 83.3% xAI releases Grok 4.5 with low cost and mixed benchmark performance
2026-07-07 LLM-as-a-Verifier 86.5% LLM-as-a-Verifier introduces general-purpose verification framework with continuous scoring
2026-07-05 GLM 5.2 FP8 with FP8 KV 79.8% GLM 5.2 FP8 with FP8 KV achieves 79.8% on Terminal-Bench 2.1
2026-06-23 Tmax-27B 43.0% Tmax-27B Terminal Agent for Small GPUs with DPPO Training
2026-06-23 Tmax 27.0% Tmax: A Simple RL Recipe for Terminal Agents
2026-04-25 Qwen3.6-27B 59.3% Alibaba releases Qwen3.6-27B, beating larger predecessor on coding benchmarks
2026-04-25 GPT-5.5 82.7% OpenAI releases GPT-5.5, a ground-up retrained model with native omnimodal processing
2026-04-23 GPT-5.5 82.7% OpenAI releases GPT-5.5 with agentic capabilities and doubled API pricing
2026-04-23 GPT-5.5 82.7% OpenAI releases GPT-5.5 and GPT-5.5 Pro in API
2026-04-12 MiniMax M2.7 57.0% MiniMax open-sources M2.7, a self-evolving MoE model scoring 56.22% on SWE-Pro
2026-02-10 Step 3.5 Flash 51.0% Step 3.5 Flash: open 11B-active MoE model achieves frontier-level agentic performance
2025-11-19 GPT-5.1-Codex-Max 58.1% OpenAI makes GPT-5.1-Codex-Max the default in Codex CLI