FLAMES
Models respond to Chinese prompts about safety and values. Evaluators assess the responses against the benchmark's alignment criteria.
- Made by
- Shanghai AI Laboratory and collaborators
- Released
- November 2023
- Kind
- Deception and alignment
- Level
- Everyday
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Propensity
- Risk
- Hate, Non-violent crimes, Violent crimes, Privacy
- Behavior
- Harmful use, Bias, Misaligned with values
- Language
- Chinese
- Interaction
- Single turn
- Items from
- Written by experts
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Content safety
- Strongest model tested
-
Qwen-14B #214 on the capability index
- Last model added
- October 2023
- Scores refreshed
- November 20th, 2026
Claude (snapshot unspecified) tops the leaderboard with 63.8% Baichuan2-13B-Chat leads with 43.2% among models released by October 2023
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Baichuan2-13B-Chat, at 43.2%. This view shows only models we track, each at its best.
The newest model with a score here came out in October 2023, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 63.8 Claude (snapshot unspecified)
- 58.6 InternLM-20B-Chat
- 53.9 InternLM-7B-Chat
- 46.9 ChatGPT
- 46.2 Baichuan2-7B-Chat
- 46.0 Ernie Bot
- 43.2 Baichuan2-13B-Chat
- 42.0
Qwen-14B
- 41.1 ChatYuan-large-v2
- 40.0 GPT-4 (snapshot unspecified)
- 36.5 Qwen-7B-Chat
- 36.3 ChatGLM3-6B
- 36.2 MOSS
- 33.9 chatglm2-6b
- 33.1 ChatGLM-6B
- 24.8 BELLE
- 23.7 MiniMax
- 1 Claude (snapshot unspecified) 63.8%
- 2 InternLM-20B-Chat 58.6%
- 3 InternLM-7B-Chat 53.9%
- 4 ChatGPT 46.9%
- 5 Baichuan2-7B-Chat 46.2%
- 6 Ernie Bot 46%
- 7 Baichuan2-13B-Chat 43.2%
- 8
Qwen-14B 42%
- 9 ChatYuan-large-v2 41.1%
- 10 GPT-4 (snapshot unspecified) 40%
- 11 Qwen-7B-Chat 36.5%
- 12 ChatGLM3-6B 36.3%
- 13 MOSS 36.2%
- 14 chatglm2-6b 33.9%
- 15 ChatGLM-6B 33.1%
- 16 BELLE 24.8%
- 17 MiniMax 23.7%
Source: Official FLAMES leaderboard
- 1 Baichuan2-13B-Chat 43.2%
- 2
Qwen-14B 42%
- 3 ChatGLM3-6B 36.3%
- 4 chatglm2-6b 33.9%
- 5 ChatGLM-6B 33.1%
Source: Official FLAMES leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude (snapshot unspecified) closed weights | 63.8% | Official leaderboard | ||
| 2 | InternLM-20B-Chat open weights, 20B | 58.6% | Official leaderboard | ||
| 3 | InternLM-7B-Chat open weights, 7B | 53.9% | Official leaderboard | ||
| 4 | ChatGPT closed weights | 46.9% | Official leaderboard | ||
| 5 | Baichuan2-7B-Chat open weights, 7B | 46.2% | Official leaderboard | ||
| 6 | Ernie Bot closed weights | 46% | Official leaderboard | ||
| 7 | Baichuan2-13B-Chat | 43.2% | Sep 2023 | Official leaderboard | |
| 8 | | 42% | Sep 2023 | Official leaderboard | |
| 9 | ChatYuan-large-v2 open weights, 7B | 41.1% | Official leaderboard | ||
| 10 | GPT-4 (snapshot unspecified) closed weights | 40% | Official leaderboard | ||
| 11 | Qwen-7B-Chat open weights, 7B | 36.5% | Official leaderboard | ||
| 12 | ChatGLM3-6B | 36.3% | Oct 2023 | Official leaderboard | |
| 13 | MOSS open weights, 16B | 36.2% | Official leaderboard | ||
| 14 | chatglm2-6b | 33.9% | Jun 2023 | Official leaderboard | |
| 15 | ChatGLM-6B | 33.1% | Mar 2023 | Official leaderboard | |
| 16 | BELLE open weights, 13B | 24.8% | Official leaderboard | ||
| 17 | MiniMax closed weights | 23.7% | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Baichuan2-13B-Chat | 43.2% | Sep 2023 | Official leaderboard | |
| 2 | | 42% | Sep 2023 | Official leaderboard | |
| 3 | ChatGLM3-6B | 36.3% | Oct 2023 | Official leaderboard | |
| 4 | chatglm2-6b | 33.9% | Jun 2023 | Official leaderboard | |
| 5 | ChatGLM-6B | 33.1% | Mar 2023 | Official leaderboard |
Rankings by part
We read 5 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Fairness, Safety and Morality. Baichuan2-13B-Chat leads 2 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Fairness
5 models we track have a score here.
- 1 Baichuan2-13B-Chat 38.5%
- 2 ChatGLM3-6B ↑1 37.8%
- 3 chatglm2-6b ↑1 31.7%
- 4
Qwen-14B ↓2 30.9%
- 5 ChatGLM-6B 26.9%
Safety
5 models we track have a score here.
- 1 Baichuan2-13B-Chat 53.9%
- 2
Qwen-14B 36.8%
- 3 ChatGLM3-6B 32.6%
- 4 chatglm2-6b 22.6%
- 5 ChatGLM-6B 15.4%
Morality
5 models we track have a score here.
- 1
Qwen-14B ↑1 54.2%
- 2 Baichuan2-13B-Chat ↓1 44.8%
- 3 ChatGLM3-6B 44.8%
- 4 chatglm2-6b 43.3%
- 5 ChatGLM-6B 40.3%
Legality
5 models we track have a score here.
- 1 ChatGLM-6B ↑4 50%
- 2 Baichuan2-13B-Chat ↓1 39.1%
- 3
Qwen-14B ↓1 32.6%
- 4 chatglm2-6b 28.3%
- 5 ChatGLM3-6B ↓2 28.3%
Data protection
5 models we track have a score here.
- 1
Qwen-14B ↑1 55.3%
- 2 chatglm2-6b ↑2 43.4%
- 3 Baichuan2-13B-Chat ↓2 39.5%
- 4 ChatGLM3-6B ↓1 38.2%
- 5 ChatGLM-6B 32.9%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official FLAMES leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official FLAMES leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More 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%
- 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.
Gemini 4 Carbon 0%
- Covert misbehavior (Thinking Machines) In use Agentic tasks that check whether a model misbehaves and conceals it. Scored by the rate of covert misbehavior.
Inkling 2 Mini 8.30%
- Destructive actions (Thinking Machines) In use Agentic tasks that check whether a model takes destructive actions it was not authorized to take. Scored by the rate of destructive actions.
Inkling 2 Small 4.70%
- Misbehavior (Thinking Machines) In use Agentic tasks that check whether a model fabricates results, ignores instructions or hides what it did when a task is hard. Scored by the rate of misbehavior.
Inkling 2 Mini 9.80%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 3 years ago, in October 2023, for ChatGLM3-6B. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the official leaderboard (17 entries).
- Pass Models scored 5 models we track, plus 12 other entries on its leaderboard.
- Warning Newest score The newest came 3 years ago, in October 2023, for ChatGLM3-6B.
- Note Frontier models None of the top 10 has a score. The strongest tested is Qwen-14B (#214).
- Pass Headroom The best is 43.2%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on flames.opencompass.org.cn: 5 rows for models we track and 12 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.