AI Safety Evidence

Freedom

Which models are most likely to do what the user asks?

This ranking measures user-direction following—answering or completing what the user asks instead of refusing. Refusals are treated as bad. It has one component, no company-governance score, and is separate from the Safety & Ethics ranking.

Safety-oriented refusal endpoints are reversed relative to the main ranking. Completion, benign-answer, and over-refusal endpoints are oriented directly so that more answering and less refusal are always better.

Freedom ranking

185 models have at least 1 qualifying evaluation. This view includes only verified releases from 2025-09-15 through 2026-09-15; ranks remain relative to the full 645-model fitted portfolio. Brackets show 90% source-sensitivity intervals; the median interval spans 289 ranks.

Freedom and intelligence

Among 147 recent models with a reviewed Artificial Analysis match, this static chart compares the pinned AA Intelligence Index with absolute Freedom rank. Rank 1 is best.

A small green check at a developer mark's lower right identifies a base model with at least one pinned direct refusal-removal finetune. The x-axis uses the highest published Intelligence Index v4.3 across matched configurations in the pinned 2026-09-15 AA catalog, without mixing older index versions. Dashed outlines identify AA-estimated intelligence values; these are also flagged in the download and point descriptions. Current plot-only AA values are not fed back into the fit. Download plotted data.

185 models
Freedom rank; brackets show the 90% source-sensitivity interval.
Estimated rank
[90% interval]		ModelWeightsFinetunesEvalsRelease date
2 [4–63]Mistral Large 3🤗—62025-12-01
11 [18–127]Ministral 3 8B🤗32025-10-31
17 [30–114]Mistral Medium 3.5🤗darkc0dedarkc0de32026-03-31
21 [30–114]GLM 4.7 Flash🤗huihui-ai32026-01-19
27 [36–201]DeepSeek v3.2 Speciale🤗—32025-11-28
36 [37–241]Gemini 3.6 Flash——32026-07-21
41 [27–357]Mistral Small 4🤗darkc0dedealignaidealignaidealignai42026-01-23
47 [32–501]Ministral 3 14B🤗22025-10-31
49 [3–408]Devstral 2——12025-12-09
53 [67–290]DeepSeek v3.1 Terminus🤗—32025-09-22
55 [21–525]Qwen3-VL 235B A22B Instruct🤗22025-09-22
59 [68–274]DeepSeek v3.2 Exp🤗—42025-09-29
63 [80–230]GLM 4.6🤗52025-09-30
72 [35–326]Grok 4.5——62026-07-08
75 [42–266]GLM 5 Turbo——42026-03-15
76 [44–479]Ministral 3 3B🤗22025-10-31
84 [34–348]Trinity Large🤗—32026-01-27
91 [40–450]Grok 4.6——32026-08-12
101 [55–401]Gemini 3.5 Flash Lite——32026-07-21
115 [85–423]Nemotron 3.5 Lightning🤗22026-08-11
120 [10–564]Qwen3-VL 8B Instruct🤗heretic-orghuihui-aihuihui-ai12025-10-11
129 [72–328]GLM 4.7🤗—52026-06-03
132 [1–586]Qwen3-VL 4B Instruct🤗huihui-ai12025-10-11
139 [10–558]LFM2 24B A2B🤗huihui-ai12026-02-24
142 [98–426]Claude Mythos 5.1——12026-09-01
154 [36–502]LFM2 8B A1B🤗huihui-ai12025-10-07
163 [25–477]Grok 4.20 Multi Agent——22026-03-10
188 [75–484]ERNIE 5.0 Thinking——12026-01-22
190 [114–419]GPT 5.5 Pro——22026-04-23
193 [75–482]Qwen3.5 9B🤗Blackfrost-AIdealignaihuihui-ai32026-02-27
195 [5–562]Solar Pro 4——12026-08-11
200 [105–330]DeepSeek v3.2🤗—112025-12-01
209 [59–454]DeepSeek v4 Flash🤗—62026-04-22
210 [88–553]Qwen3-VL 30B A3B🤗dealignaidealignaihuihui-ai12025-09-30
214 [115–367]Gemini 3.1 Flash Lite——92026-03-03
217 [69–510]Ling 1T——12025-10-02
227 [79–460]Ling 2.5 1T🤗—12026-02-14
228 [157–392]Gemini 3 Pro Preview——142025-11-18
229 [52–536]Muse Spark 1.2——12026-08-05
230 [110–435]GLM 5.1🤗—62026-04-03
232 [69–555]Qwen3-VL 32B Instruct🤗huihui-ai12025-10-19
233 [160–380]Gemma 4 E2B——12026-04-02
234 [136–399]Gemini 3.5 Flash——92026-05-19
239 [120–432]Doubao Seed 1.8——12025-12-15
246 [141–422]GPT 5.2 Chat——12025-12-11
247 [142–420]Nemotron 3 Nano 30B A3B🤗—52025-12-15
254 [119–403]Hy3 Preview🤗—32026-04-13
256 [117–431]MiMo v2 Flash🤗—52025-12-16
264 [105–492]Ernie 5.0 Thinking Preview——12025-11-13
265 [110–481]DeepSeek V4 Flash Vision——12026-08-21
266 [106–464]Kimi K3🤗52026-07-16
269 [99–463]Grok 4.1 Fast——122025-11-19
270 [73–643]Muse Glimmer🤗Blackfrost-AI32026-08-10
277 [143–439]DeepSeek v4 Pro🤗—82026-04-22
279 [116–394]GLM 4.6V🤗—22025-12-07
287 [111–567]Qwen3.5 2B🤗huihui-ai12026-02-28
288 [69–617]Ring 1T——12025-10-10
291 [116–456]Ernie 5.0 Preview——12025-10-22
300 [106–527]GLM 5.3🤗32026-08-18
306 [90–480]Longcat 2.0🤗—32026-07-05
319 [153–478]Ernie 5.1——12026-05-09
323 [143–492]Nanbeige4.1 3B🤗heretic-org12026-02-10
329 [94–617]Qwen3.5 0.8B🤗darkc0dehuihui-ai12026-02-28
330 [161–483]Nemotron 3 Ultra 550B A55B——32026-06-03
331 [172–463]Grok 4.20——92026-03-10
332 [179–471]Doubao Seed 2.0 Pro——22026-02-14
333 [227–434]Gemma 4 E4B——22026-03-31
339 [154–504]OLMo 3 7B Think🤗22025-11-18
340 [144–503]Gemini 3 Flash Preview——82025-12-17
341 [162–454]Gemini 3.1 Pro Preview——132026-02-19
348 [178–492]Gemma 4 26B A4B IT🤗huihui-aihuihui-aihuihui-ai52026-03-11
358 [182–514]Intellect 3🤗—12025-11-26
362 [191–496]Gemma 4 31B IT🤗huihui-aihuihui-ai72026-03-11
373 [198–491]Apriel v1.6 15B Thinker🤗—12025-11-28
375 [236–474]Hunyuan 2.0 Think——12025-11-09
376 [164–533]Grok 4 Fast——62025-09-19
381 [186–475]GPT 5.3 Codex——12026-02-05
383 [179–572]Qwen3.6 27B🤗darkc0dedealignaidealignaidealignaihuihui-ai12026-04-21
386 [252–521]Gemma 4 12B——12026-06-03
387 [225–514]Laguna S 2.1 Poolside🤗dealignaidealignaidealignai22026-07-21
391 [187–510]Step 3.5 Flash🤗32026-02-01
393 [253–480]GLM 5.3 Flash🤗orcarouter22026-08-26
396 [251–538]Qwen3.8 Max🤗—22026-08-03
398 [207–538]GPT OSS Safeguard 20B🤗—22025-10-29
401 [203–608]Granite 4.0 Micro🤗huihui-ai62025-10-02
403 [120–591]Qwen3.5 27B🤗darkc0dedarkc0dedarkc0dedarkc0dedarkc0dedealignaihuihui-ai52026-02-24
406 [212–540]Gemini 3.7 Flash——22026-08-13
415 [164–636]Qwen3.5 4B🤗darkc0dedealignai22026-02-27
416 [273–533]MiniMax M2.1🤗—42025-12-20
419 [264–587]Ring 2.6 1T——12026-05-14
420 [105–579]MiMo v2.5🤗—42026-04-27
421 [272–507]Kimi K2.5🤗132026-01-01
422 [272–506]Gemini 3.1 Pro——12026-02-19
425 [283–563]GLM 5🤗72026-02-11
430 [293–508]GLM 5V Turbo——12026-04-01
433 [38–620]GPT 6 Astra——22026-09-03
438 [294–570]GLM 5.2🤗—92026-06-16
439 [288–555]Granite 4.2 8B——12026-08-25
442 [246–542]Qwen3 Max——52025-09-24
450 [311–545]Mercury 2——22026-02-24
451 [290–530]Healer Alpha——12026-03-11
453 [254–551]DeepSeek V4 Flash 0731🤗Blackfrost-AIdealignaidealignai12026-07-31
455 [314–571]Qwen3.7 Max——62026-05-20
457 [291–566]Gemini 3.8 Flash——22026-09-02
458 [110–634]MiMo v2 Omni——22026-03-18
463 [299–544]Granite 4.0 Small🤗—62025-10-02
465 [312–559]OLMo 3 32B Think🤗—22025-11-19
466 [131–635]Inkling Small🤗—22026-07-30
469 [276–591]Nova 2.0 Lite——12025-12-02
472 [277–556]Ling 2.6 1T——12026-04-29
473 [281–581]Grok 4.3——102026-05-15
476 [316–568]Ox Alpha——12026-08-20
478 [317–561]Qwen3-VL 8B🤗huihui-ai12025-10-11
481 [316–591]Seed 2.0 Lite——12026-02-14
482 [309–595]Kimi K2.6🤗92026-04-14
484 [201–636]Nemotron 3 Nano Omni 30B A3B——22026-04-28
486 [337–575]Intern S1 Pro🤗—12026-02-02
488 [359–522]GPT 5.3 Instant——12026-03-03
489 [257–577]Ling 3.0 Flash🤗Blackfrost-AI12026-08-02
490 [352–584]GPT 5.1——172025-11-13
491 [252–567]Qwen3.5 35B A3B🤗huihui-ai32026-02-24
493 [111–645]MiMo v2 Pro——32026-03-18
494 [315–566]GPT-OSS Safeguard 120B🤗—12025-09-18
499 [366–545]MiniMax M2🤗—52025-10-27
502 [310–580]Qwen3.5 397B A17B🤗—92026-02-16
503 [301–600]MiMo v2.5 Pro🤗—52026-04-27
506 [344–572]GPT 5.6 Sol——82026-07-09
508 [327–593]Qwen3-VL 235B A22B🤗—12025-09-22
513 [231–600]Cogito v2.1——12025-11-19
514 [332–608]MiniMax M2.5🤗dealignaidealignaidealignai52026-02-12
515 [349–572]Muse Spark——12026-04-08
516 [294–596]Nemotron 3 Ultra——12026-06-04
517 [157–642]Seed 2.0 Mini——22026-02-14
519 [325–614]Nemotron 3 Super 120B A12B——52026-03-10
520 [398–613]Sarvam 30B——12026-03-03
521 [200–632]Muse Spark 1.1——52026-07-09
526 [340–564]GPT 5.2 Instant——12025-12-11
527 [357–617]Solar Pro 3——12026-01-27
531 [303–642]MiniMax M3🤗32026-06-02
532 [337–575]GPT 5.6 Luna——92026-07-09
534 [321–624]GPT 5.3 Chat——42026-03-03
536 [360–598]Qwen3.6 35B A3B🤗dealignaidealignaihuihui-ai12026-04-15
543 [358–610]GPT 5.5——152026-04-23
544 [336–615]Qwen3.5 122B A10B🤗dealignaidealignai32026-02-24
547 [370–613]Step 3.7 Flash🤗12026-05-23
556 [356–622]Qwen3.7 Plus——22026-05-26
560 [354–624]Claude Opus 4.7——112026-04-16
561 [264–636]Apriel v1.5 15B Thinker🤗—12025-09-24
562 [386–573]Granite 4.0 H Micro🤗huihui-ai12025-09-16
565 [360–611]GPT 5.6 Terra——72026-07-09
569 [308–643]Inkling🤗—32026-07-14
571 [383–598]Qwen3.5 Flash——32026-02-24
574 [361–609]Granite 4.0 H Tiny——12025-09-16
585 [385–605]Granite 4.0 H Small——12025-10-02
587 [320–640]Aurora Alpha——22026-02-09
589 [396–612]Claude Haiku 4.5——222025-10-15
591 [312–645]Hunter Alpha——12026-03-11
595 [340–633]OLMo 3 7B Instruct🤗—22025-11-19
597 [451–613]MiniMax M2.7🤗dealignaidealignaidealignaidealignai72026-04-09
598 [332–642]GPT 5.5 Instant——12026-05-26
600 [339–636]GPT 5.4 Pro——12026-03-05
604 [358–644]Ling 2.6 Flash——12026-04-21
605 [365–641]GPT 5.4 Nano——62026-03-17
608 [377–640]Qwen3.6 Plus——52026-04-02
609 [331–643]Claude Opus 5——42026-07-24
611 [397–636]Qwen3.8 27B🤗Blackfrost-AIheretic-orghuihui-aiorcarouter12026-08-14
616 [392–631]GPT 5.1 Chat——22025-11-13
618 [438–635]Qwen3.5 Plus——52026-02-15
619 [476–624]Claude Sonnet 4.5——212025-09-29
620 [466–621]GPT 5.2——152025-12-11
623 [312–642]Claude Fable 5.1——42026-09-01
624 [365–628]Granite 4.0 Micro Guardian——52025-10-02
625 [419–643]OLMo 3.1 32B Think🤗—12025-12-10
629 [328–641]Hy3🤗—22026-07-02
630 [434–639]Claude Sonnet 5——72026-06-30
631 [465–637]GPT 5.4 Mini——72026-03-17
635 [483–634]Claude Sonnet 4.6——142026-01-21
636 [469–636]GPT 5.4——142026-03-05
637 [505–633]Sarvam 105B——12026-03-03
638 [404–640]Granite 4.0 Small Guardian——52025-10-02
639 [480–637]Claude Opus 4.8——112026-05-28
640 [413–644]Claude Fable 5——72026-06-09
641 [454–644]Qwen3.7 Flash——12026-07-27
642 [490–634]Claude Opus 4.6——132026-02-05
644 [491–636]Claude Opus 4.5——122025-11-24

How it is built

1 component—User direction following—contain 265 sub-evaluations from 69 governed lineages. Sibling metrics share one lineage budget. The estimator uses the same ordinal consensus and lineage-bootstrap machinery as the main ranking, but with this ranking's own source policy, component contract, directions, weights, and evidence gate.

Download the Freedom ranking CSV

Sources