Anthropic
xAI
OpenAI
Google DeepMind
[_model_slider][__benchmark_archive_v4 color_mode="light"] [__benchark_sort_preview id="19563"]
GPT-6 Astra
Leads Terminal-Bench 4.0 and Agents' Last Exam, strong at agentic coding.
Gemini Flash 3.8
Mid-pack across most boards, best showing on SnorkelLegal.
Muse Spark 1.3
Second on SnorkelLegal and Agents' Last Exam, weak on coding benchmarks.
Fable 5.1
Tops SWE-bench CLI, Terminal-Bench-Science, and Senior SWE-Bench.
GLM 5.3
First on SnorkelUnderwrite 2.0, strong across the Snorkel business suite.
DeepSeek V4 Pro
Second on SnorkelUnderwrite 2.0, mid pack on manufacturing, weak elsewhere.
Grok 4.6
Leads four of the five SnorkelWorkplace-family benchmarks it has entered.
Qwen 3.8 Max
Trails the field on most boards, no top-three finishes.
Claude Opus 5
Leads OSWorld 2.0, top three on SWE-bench CLI and Terminal-Bench-Science.
Kimi K3
Consistent mid-pack finisher, best result on SnorkelManufacturing.
Nemotron 3 Ultra 550B
Entered on only five benchmarks, ranking near the bottom on each.
Latest model releases
GPT-6 Astra
Leads Terminal-Bench 4.0 and Agents' Last Exam, strong at agentic coding.
Gemini Flash 3.8
Mid-pack across most boards, best showing on SnorkelLegal.
Muse Spark 1.3
Second on SnorkelLegal and Agents' Last Exam, weak on coding benchmarks.
Fable 5.1
Tops SWE-bench CLI, Terminal-Bench-Science, and Senior SWE-Bench.
GLM 5.3
First on SnorkelUnderwrite 2.0, strong across the Snorkel business suite.
DeepSeek V4 Pro
Second on SnorkelUnderwrite 2.0, mid pack on manufacturing, weak elsewhere.
Grok 4.6
Leads four of the five SnorkelWorkplace-family benchmarks it has entered.
Qwen 3.8 Max
Trails the field on most boards, no top-three finishes.
Claude Opus 5
Leads OSWorld 2.0, top three on SWE-bench CLI and Terminal-Bench-Science.
Kimi K3
Consistent mid-pack finisher, best result on SnorkelManufacturing.
Nemotron 3 Ultra 550B
Entered on only five benchmarks, ranking near the bottom on each.
Benchmarks for what frontier AI hasn't solved
Senior SWE-Bench
Evaluates coding agents on senior-level software engineering tasks.
Agents’ Last Exam
Evaluating AI agents on long-horizon, economically valuable professional workflows with verifiable outcomes.
Terminal-Bench 4.0
Evaluates terminal agents on continuously updated software-engineering tasks.
Senior SWE-Bench
Evaluates coding agents on senior-level software engineering tasks.
Digital Work Index
SnorkelManufacturing
Evaluates manufacturing decisions using fragmented plant-floor, engineering, and supplier evidence.
WorkplaceAgents
Evaluates AI agents on economically valuable professional work and verifiable workplace deliverables.
SWE-bench CLI
Evaluates validated multi-file code changes in open-source repositories.
SnorkelRevOps
Evaluates revenue-operations workflows involving systems, controls, and business-state updates.
Continual Learning Bench
Measures improvement across sequential, stateful tasks.
SnorkelFinance 2.0
Evaluates financial workflows requiring evidence gathering, analysis, and compliance.
SlopCode Bench
Measures coding-agent performance across evolving software requirements.
LibraryDesignBench
Measures how well AI agents design libraries that other agents can use.
SnorkelUnderwrite 2.0
Evaluates auditable insurance underwriting decisions using incomplete, distributed evidence.
OSWorld 2.0
Evaluates computer-use agents on 108 long-horizon workflows.
Agents’ Last Exam
Evaluating AI agents on long-horizon, economically valuable professional workflows with verifiable outcomes.
SnorkelLegal
Evaluates legal workflows requiring evidence, procedure, authority, and action.
Terminal-Bench-Science
Evaluates agents on scientific workflows derived from researchers’ own work.
Terminal-Bench 3.0
Evaluates terminal agents on containerized tasks across seven domains.
Agentic Coding 2.0
Evaluating whether coding agents can plan, execute, verify, and recover across complex terminal-native engineering tasks.
Terminal-Bench 4.0
Evaluates terminal agents on continuously updated software-engineering tasks.