# Spiral-Bench v1.2 > Source: Benchize (https://benchize.com) — https://benchize.com/benchmarks/eq-bench%3Aspiral-bench-v1.2. When citing or reusing this content, credit "Benchize" and link to https://benchize.com/benchmarks/eq-bench%3Aspiral-bench-v1.2. Safety and social handling of simulated twenty-turn conversations with users vulnerable to delusional thinking. An ensemble of judges grades protective and risky behaviors; v1.2 uses a revised rubric. Canonical URL: https://benchize.com/benchmarks/eq-bench%3Aspiral-bench-v1.2 Last updated: Wed, 21 Oct 2026 18:52:20 GMT Reproduction and quotation must name Benchize, link to the canonical HTML URL, and retain individual source citations. Safety and social handling of simulated twenty-turn conversations with users vulnerable to delusional thinking. An ensemble of judges grades protective and risky behaviors; v1.2 uses a revised rubric. Safety & robustness · Models tested: 16 · Score publishers: [eq bench](https://eqbench.com/spiral-bench.html) ## Model results | # | Model | Score | ± | | | --- | --- | --- | --- | --- | | 1 | [Claude Sonnet 4.5](https://benchize.com/models/claude-sonnet-4-5.md) | 70.3% | CI 70.3–70.3% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 70.3 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion claude-sonnet-4.5 ### Reported statistics Uncertainty Confidence interval: 70.3–70.3% confidenceIntervalHigh 70.3 confidenceIntervalLow 70.3 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label claude-sonnet-4.5 Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/2` | | | | | | 2 | [GPT-5.2](https://benchize.com/models/gpt-5-2.md) | 70.2% | CI 70.2–70.2% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 70.2 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion gpt-5.2 ### Reported statistics Uncertainty Confidence interval: 70.2–70.2% confidenceIntervalHigh 70.2 confidenceIntervalLow 70.2 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label gpt-5.2 Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/21` | | | | | | 3 | [O3](https://benchize.com/models/o3.md) | 63.4% | CI 63.4–63.4% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 63.4 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion o3 ### Reported statistics Uncertainty Confidence interval: 63.4–63.4% confidenceIntervalHigh 63.4 confidenceIntervalLow 63.4 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label o3 Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/16` | | | | | | 4 | [gpt-oss-120b](https://benchize.com/models/gpt-oss-120b.md) | 57.5% | CI 57.5–57.5% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 57.5 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion gpt-oss-120b ### Reported statistics Uncertainty Confidence interval: 57.5–57.5% confidenceIntervalHigh 57.5 confidenceIntervalLow 57.5 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label gpt-oss-120b Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/11` | | | | | | 5 | [Kimi K2](https://benchize.com/models/kimi-k2.md) | 57.2% | CI 57.2–57.2% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 57.2 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion kimi-k2 ### Reported statistics Uncertainty Confidence interval: 57.2–57.2% confidenceIntervalHigh 57.2 confidenceIntervalLow 57.2 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label kimi-k2 Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/13` | | | | | | 6 | [O4 Mini](https://benchize.com/models/o4-mini.md) | 53.6% | CI 53.6–53.6% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 53.6 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion o4-mini ### Reported statistics Uncertainty Confidence interval: 53.6–53.6% confidenceIntervalHigh 53.6 confidenceIntervalLow 53.6 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label o4-mini Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/17` | | | | | | 7 | [Claude 3.5 Sonnet](https://benchize.com/models/claude-3-5-sonnet.md) | 42.6% | CI 42.6–42.6% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 42.6 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion claude-3.5-sonnet ### Reported statistics Uncertainty Confidence interval: 42.6–42.6% confidenceIntervalHigh 42.6 confidenceIntervalLow 42.6 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label claude-3.5-sonnet Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/1` | | | | | | 8 | [Llama 4 Maverick](https://benchize.com/models/llama-4-maverick.md) | 38.2% | CI 38.2–38.2% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 38.2 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion llama-4-maverick ### Reported statistics Uncertainty Confidence interval: 38.2–38.2% confidenceIntervalHigh 38.2 confidenceIntervalLow 38.2 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label llama-4-maverick Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/14` | | | | | | 9 | [Gemini 2.5 Flash](https://benchize.com/models/gemini-2-5-flash.md) | 36.5% | CI 36.5–36.5% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 36.5 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion gemini-2.5-flash ### Reported statistics Uncertainty Confidence interval: 36.5–36.5% confidenceIntervalHigh 36.5 confidenceIntervalLow 36.5 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label gemini-2.5-flash Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/5` | | | | | | 10 | [Claude Sonnet 4](https://benchize.com/models/claude-sonnet-4.md) | 32.1% | CI 32.1–32.1% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 32.1 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion claude-sonnet-4 ### Reported statistics Uncertainty Confidence interval: 32.1–32.1% confidenceIntervalHigh 32.1 confidenceIntervalLow 32.1 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label claude-sonnet-4 Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/3` | | | | | | 11 | [GLM-4.6](https://benchize.com/models/glm-4-6.md) | 30.8% | CI 30.8–30.8% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 30.8 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion glm-4.6 ### Reported statistics Uncertainty Confidence interval: 30.8–30.8% confidenceIntervalHigh 30.8 confidenceIntervalLow 30.8 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label glm-4.6 Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/7` | | | | | | 12 | [Gemini 2.5 Pro](https://benchize.com/models/gemini-2-5-pro.md) | 28.3% | CI 28.3–28.3% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 28.3 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion gemini-2.5-pro ### Reported statistics Uncertainty Confidence interval: 28.3–28.3% confidenceIntervalHigh 28.3 confidenceIntervalLow 28.3 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label gemini-2.5-pro Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/6` | | | | | | 13 | [Gemma 3 27B](https://benchize.com/models/gemma-3-27b-it.md) | 25.2% | CI 25.2–25.2% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 25.2 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion gemma-3-27b-it ### Reported statistics Uncertainty Confidence interval: 25.2–25.2% confidenceIntervalHigh 25.2 confidenceIntervalLow 25.2 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label gemma-3-27b-it Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/22` | | | | | | 14 | [Qwen3 235B A22B Instruct 2507](https://benchize.com/models/qwen3-235b-a22b-2507.md) | 20.4% | CI 20.4–20.4% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 20.4 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion qwen3-235b-a22b-2507 ### Reported statistics Uncertainty Confidence interval: 20.4–20.4% confidenceIntervalHigh 20.4 confidenceIntervalLow 20.4 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label qwen3-235b-a22b-2507 Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/18` | | | | | | 15 | [Mistral Medium 3.1](https://benchize.com/models/mistral-medium-3-1.md) | 15% | CI 15–15% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 15.0 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion mistral-medium-3.1 ### Reported statistics Uncertainty Confidence interval: 15–15% confidenceIntervalHigh 15.0 confidenceIntervalLow 15.0 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label mistral-medium-3.1 Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/15` | | | | | | 16 | [R1 0528](https://benchize.com/models/deepseek-r1-0528.md) | 14.2% | CI 14.2–14.2% | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score_norm` \| 14.2 \| percent \| overall \| Higher is better \| ### Configuration interactiveAdjustments none protocol spiral-bench-v1.2 sourceModelVersion deepseek-r1-0528 ### Reported statistics Uncertainty Confidence interval: 14.2–14.2% confidenceIntervalHigh 14.2 confidenceIntervalLow 14.2 ### Source [Original source](https://eqbench.com/spiral-bench.js) Source eq bench Original model label deepseek-r1-0528 Source version spiral-bench-v1.2 Observed 2026-10-21T16:35:04.308Z Evidence SHA-256 `2f0402654b01e7363c0d02bbc708afd7849e669dc5da63d9f85c995ed8f0a97f` Source locator `/leaderboardDataDelusion/4` | | | | | --- Cite as: Spiral-Bench v1.2 — Benchize, https://benchize.com/benchmarks/eq-bench%3Aspiral-bench-v1.2 (retrieved 2026-10-21)