BlogLLM Benchmark Scores as Published, Side by Side
Seal-0: Results as Published
Every result the catalog holds for Seal-0: 17 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 Thinh Pham and co-authors (Virginia Tech)Introduced 2025Data version 2026103001, as of 2026-10-30
Seal-0 is the main subset of SealQA, a set of fact-seeking questions whose web search returns conflicting, noisy, or unhelpful pages. It has 111 questions that were rewritten until strong non-reasoning chat models with browsing still failed. The model has to answer from that messy evidence rather than from the first page. The reported result is accuracy.
Seal-0
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-5.2OpenAI, released 2025-12-11 | 45% | Alibaba, Qwen3.5-397B-A17B model cardchecked 2026-10-15; printed on 1 more page | Tools: enabled; search; Context: 256000 |
| GPT-5OpenAI, released 2025-08-07 | 51.4% | Moonshot AI, Kimi K2 Thinking release pagechecked 2026-10-15 | Effort: high; Tools: enabled; search, code interpreter, web browsing; Context: 256000 |
| GPT-5 miniOpenAI, released 2025-08-07 | 34.2% | Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15 | Tools: enabled; search; Context: 256000 |
| gpt-oss-120bOpenAI, released 2025-08-05 | 45.1% | Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15 | Tools: enabled; search; Context: 256000 |
| Claude Opus 4.5Anthropic, released 2025-11-24 | 47.7% | Alibaba, Qwen3.5-397B-A17B model cardchecked 2026-10-15; printed on 1 more page | Tools: enabled; search; Context: 256000 |
| Claude Sonnet 4.5Anthropic, released 2025-09-29 | 53.4% | Moonshot AI, Kimi K2 Thinking release pagechecked 2026-10-15 | Effort: thinking; Tools: enabled; search, code interpreter, web browsing; Context: 256000 |
| Gemini 3 ProGoogle, released 2025-11-18 | 45.5% | Alibaba, Qwen3.5-397B-A17B model cardchecked 2026-10-15; printed on 1 more page | Tools: enabled; search; Context: 256000 |
| DeepSeek-V3.2DeepSeek, released 2025-12-01 | 49.5% | Moonshot AI, Kimi K2.5 model cardchecked 2026-10-15 | Tools: search, code interpreter, browsing |
| DeepSeek-V3.2-ExpDeepSeek, released 2025-09-29 | 38.5% | Moonshot AI, Kimi K2 Thinking release pagechecked 2026-10-15 | Tools: enabled; search, code interpreter, web browsing; Context: 256000 |
| Qwen3.5-122B-A10BAlibaba, released 2026-02-24 | 44.1% | Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15 | Tools: enabled; search; Context: 256000 |
| Qwen3.5-27BAlibaba, released 2026-02-24 | 47.2% | Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15 | Tools: enabled; search; Context: 256000 |
| Qwen3.5-35B-A3BAlibaba, released 2026-02-24 | 41.4% | Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15 | Tools: enabled; search; Context: 256000 |
| Qwen3.5-397B-A17BAlibaba, released 2026-02-16 | 46.9% | Alibaba, Qwen3.5-397B-A17B model cardchecked 2026-10-15 | Tools: enabled; search; Context: 256000 |
| Qwen3-Max-ThinkingAlibaba | 46.9% | Alibaba, Qwen3.5-397B-A17B model cardchecked 2026-10-15 | Tools: enabled; search; Context: 256000 |
| Kimi K2.5Moonshot AI, released 2026-01-27 | 57.4% | Moonshot AI, Kimi K2.5 model cardchecked 2026-10-15; printed on 1 more page | Tools: search, code interpreter, browsing |
| Kimi K2 ThinkingMoonshot AI, released 2025-11-06 | 56.3% | Moonshot AI, Kimi K2 Thinking release pagechecked 2026-10-15 | Effort: thinking; Tools: enabled; search, code interpreter, web browsing; Context: 256000 |
| Kimi K2 Instruct 0905Moonshot AI, released 2025-09-05 | 25.2% | Moonshot AI, Kimi K2 Thinking release pagechecked 2026-10-15 | Tools: enabled; search, code interpreter, web browsing; Context: 256000 |
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.