BlogLLM Benchmark Scores as Published, Side by Side

FRAMES: Results as Published

Every result the catalog holds for FRAMES: 13 models from 6 developers, read from 4 pages. Each number is as printed by the company named beside it; nothing is ranked or rescored. Models are listed by developer, newest first.

Area Web Search and ResearchMaintainer Google (Satyapriya Krishna and co-authors)Introduced 2024Data version 2026103001, as of 2026-10-30

FRAMES (Fact, Fetch, and Reason) tests whether a model can answer multi-hop questions that need facts from several web sources combined by reasoning. It is used to evaluate retrieval-augmented and search-agent setups. The score is the share of questions answered correctly.

FRAMES

ModelValuePrinted bySetup as printed
GPT-5OpenAI, released 2025-08-0786%Moonshot AI, Kimi K2 Thinking release pagechecked 2026-10-15; printed on 1 more pageEffort: high; Tools: enabled; search, code interpreter, web browsing; Context: 256000
o1-miniOpenAI, released 2024-09-1276.9%DeepSeek, DeepSeek-R1 model cardchecked 2026-10-15-
GPT-4oOpenAI, released 2024-05-1380.5%DeepSeek, DeepSeek-R1 model cardchecked 2026-10-15-
Claude Sonnet 4.5Anthropic, released 2025-09-2985%Moonshot AI, Kimi K2 Thinking release pagechecked 2026-10-15; printed on 1 more pageEffort: thinking; Tools: enabled; search, code interpreter, web browsing; Context: 256000
Claude 3.5 SonnetAnthropic, released 2024-10-2272.5%DeepSeek, DeepSeek-R1 model cardchecked 2026-10-15-
Gemini 3 ProGoogle, released 2025-11-1890.9%xAI, Grok 4.1 Fast and Agent Tools APIchecked 2026-10-15Tools: agentic search (web search, X search, code execution); exact tools per row not stated
Grok 4.1 FastxAI, released 2025-11-1987.6%xAI, Grok 4.1 Fast and Agent Tools APIchecked 2026-10-15Tools: agentic search (web search, X search, code execution); exact tools per row not stated
DeepSeek-V3.2-ExpDeepSeek, released 2025-09-2980.2%Moonshot AI, Kimi K2 Thinking release pagechecked 2026-10-15Tools: enabled; search, code interpreter, web browsing; Context: 256000
DeepSeek-R1-0528DeepSeek, released 2025-05-2883%DeepSeek, DeepSeek-R1-0528 model cardchecked 2026-10-15-
DeepSeek-R1DeepSeek, released 2025-01-2082.5%DeepSeek, DeepSeek-R1 model cardchecked 2026-10-15; printed on 1 more page-
DeepSeek-V3DeepSeek, released 2024-12-2673.3%DeepSeek, DeepSeek-R1 model cardchecked 2026-10-15-
Kimi K2 ThinkingMoonshot AI, released 2025-11-0687%Moonshot AI, Kimi K2 Thinking release pagechecked 2026-10-15Effort: thinking; Tools: enabled; search, code interpreter, web browsing; Context: 256000
Kimi K2 Instruct 0905Moonshot AI, released 2025-09-0558.1%Moonshot AI, Kimi K2 Thinking release pagechecked 2026-10-15Tools: enabled; search, code interpreter, web browsing; Context: 256000

FRAMES (average cost per task)

lower is better

ModelValuePrinted bySetup as printed
GPT-5OpenAI, released 2025-08-07$0.058xAI, Grok 4.1 Fast and Agent Tools APIchecked 2026-10-15Tools: agentic search (web search, X search, code execution); exact tools per row not stated
Claude Sonnet 4.5Anthropic, released 2025-09-29$0.078xAI, Grok 4.1 Fast and Agent Tools APIchecked 2026-10-15Tools: agentic search (web search, X search, code execution); exact tools per row not stated
Grok 4.1 FastxAI, released 2025-11-19$0.048xAI, Grok 4.1 Fast and Agent Tools APIchecked 2026-10-15Tools: agentic search (web search, X search, code execution); exact tools per row not stated

Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.