Benchmark Almanac

Broken Search Tool

GPT-6.1 Terra 1.83% leads since Oct 2026 Only in system cards

Models answer questions that need search while the search tool is broken. Scored on how often they fail to say the tool is unavailable.

In-house OpenAI evaluation. The tasks are not public.

Made by OpenAI Released Aug 2025 Deception and alignment Tasks kept private Last in a system card Oct 2026 Last model added Oct 2026
Level
Everyday Most adults, without training
Scored
Graded by a model A language model scores each response, usually against a rubric.
Answers
Agent task Many steps in a terminal, browser, desktop, app, game or robot.
7
Models scored in lab cards
10
System cards report it, from 1 lab

GPT-6.1 Terra leads with 1.83%

Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.

  1. 1 GPT-6.1 Terra 1.83%
  2. 2 GPT-6.1 Sol 2.08%
  3. 3 GPT-6 Sol 4.92%
  4. 4 GPT-6 Luna 7.83%
  5. 5 GPT-6 Astra 8.08%
  6. 6 GPT-5.2 9.10%
  7. 7 GPT-5 11%
#ModelScoreSettingDateSource
1 GPT-6.1 Terra 1.83% Oct 2026 Model card
2 GPT-6.1 Sol 2.08% Sep 2026 Model card
3 GPT-6 Sol 4.92% Sep 2026 Model card
4 GPT-6 Luna 7.83% Oct 2026 Model card
5 GPT-6 Astra 8.08% Sep 2026 Model card
6 GPT-5.2 9.10% Dec 2025 Model card
7 GPT-5 11% Aug 2025 Model card

6 models have held number one since its release in Aug 2025

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Lab claims 6 leaders
GPT-6.1 Terra 1.83%

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2026 2027 GPT-6.1 Terra GPT-6 Astra GPT-5.2 GPT-5 0% 20% 40% 60% 80% 100% 2026 2027 GPT-6.1 Terra GPT-6 Astra

Labs report it only in system cards

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark. Some system cards do, below.

10 system cards report it, from 1 lab

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
29 Oct 2026 GPT-6 Luna system card addendum 7.83 Broken Search Tool, failure rate, printed chart label (Figure 7, 7.0.4) Alignment
26 Oct 2026 GPT-6.1 Terra system card addendum 1.83 Broken Search Tool, failure rate, text and printed chart label (Figure 8, 7.1.2) Alignment
7 Oct 2026 GPT-6 Sol system card addendum 5.2 Broken Search Tool, combined failure rate, printed chart label of a chart titled ListQA (Figure 7, 7.2.2) Alignment
7 Oct 2026 GPT-6 Luna system card addendum 9.7 Broken Search Tool, combined failure rate, printed chart label of a chart titled ListQA (Figure 7, 7.2.2) Alignment
29 Sep 2026 GPT-6.1 Sol system card addendum 2.08 Broken Search Tool, fails to acknowledge unavailable search (7.4.2) Alignment
22 Sep 2026 GPT-6 Sol system card 4.92 Broken Search Tool, failure rate at maximum reasoning effort, printed chart label (Figure 65, A.6.4.2) Alignment
22 Sep 2026 GPT-6 Luna system card 28.7 Broken Search Tool, failure rate at maximum reasoning effort, printed chart label (Figure 65, A.6.4.2) Alignment
3 Sep 2026 GPT-6 Astra system card 8.08 Broken Search Tool, failure rate, printed chart label (Figure 11, 8.3.2) Alignment
11 Dec 2025 GPT-5.2 system card addendum 9.1 Deception rate, Browsing Broken Tools, gpt-5.2-thinking (Table 6) Alignment
7 Aug 2025 GPT-5 system card 11 Browsing Broken Tools, deception rate (Table 9) Alignment

More in Safety and alignment

All 444 in Safety and alignment
  • FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
  • StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is. Inkling 98.6%
  • AgentDojo In use Agents do tool-use tasks while tools and data carry prompt injections. Attack success rate counts hijacked runs. Grok 4 Fast 0%
  • MASK In use Measures whether a model contradicts its own elicited beliefs under pressure to lie. Honesty separated from accuracy. Muse Spark Open 98.1%
  • Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts. GPT-6 Astra 8.50%
  • Agentic misalignment suite (Anthropic) In use Agents in simulated company settings face threats or goal conflicts that invite blackmail, leaking or sabotage. Score is the misaligned action rate. Claude Sonnet 4.5 0%