Model profile
GPT 5.6 Sol
OpenAIdeveloper
2026-07-09release date
#30 / 312overall rank
25eval lineages
1discovery sources
Evidence summary
GPT 5.6 Sol has an estimated overall rank of #30; its 90% source-sensitivity interval is #8–#179. Its behavior-only rank is #33; company governance moves the combined estimate to #30. Published evidence spans 25 evals and 7 of 7 behavior components. Its strongest relative result is SpeciEval (belief_animal_sentience, #1 of 113); its weakest is CAIS Risk Index (textquests_harm, #50 of 50).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gpt-5-6-sol
- OpenRouter ↗openai/gpt-5.6-sol
- Official model documentation ↗Family-level model document · openai · first party
- Release source ↗direct research preview identity date
Published eval results
Rank is within that sub-eval. Black marks the observed result; the grey dot marks the value implied by the global rank. Values stay on each source’s native scale.
| Eval / sub-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #246 / 330 | ↓0.8942 | Source ↗official | |
| Arena Factuality — Search Arena (factuality-only weighting)factuality_bt_rating | #4 / 30 | ↑1241.0 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #5 / 112 | ↑1487.0 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #26 / 106 | ↑0.465 | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #3 / 45 | ↓40.3 | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #22 / 50 | ↓64.3 | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #16 / 51 | ↓46.7 | Source ↗official | |
| CAIS Risk Indexmachiavelli | #9 / 47 | ↓82.7 | Source ↗official | |
| CAIS Risk Indexmask | #3 / 53 | ↓4.6 | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #11 / 34 | ↓41.7 | Source ↗official | |
| CAIS Risk Indextextquests_harm | #50 / 50 | ↓31.4 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #50 / 241 | ↑27.91 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #195 / 241 | ↑80.83 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #6 / 241 | ↑99.44 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #194 / 239 | ↑91.09 | Source ↗official | |
| GPT-5.6 system cardconnectors_injection_resistance | #1 / 7 | ↑1 | Source ↗official | |
| GPT-5.6 system cardemotional_reliance | #4 / 7 | ↑0.953 | Source ↗official | |
| GPT-5.6 system cardextremism_not_unsafe | #6 / 7 | ↑0.962 | Source ↗official | |
| GPT-5.6 system cardgore_not_unsafe | #5 / 7 | ↑0.708 | Source ↗official | |
| GPT-5.6 system cardharm_overall_pct | #4 / 7 | ↓0.98 | Source ↗official | |
| GPT-5.6 system cardhate_not_unsafe | #4 / 7 | ↑0.982 | Source ↗official | |
| GPT-5.6 system cardmental_health | #1 / 7 | ↑0.991 | Source ↗official | |
| GPT-5.6 system cardnonviolent_illicit_not_unsafe | #6 / 7 | ↑0.987 | Source ↗official | |
| GPT-5.6 system cardsearch_function_calling_injection_resistance | #2 / 6 | ↑0.91 | Source ↗official | |
| GPT-5.6 system cardself_harm | #7 / 7 | ↑0.856 | Source ↗official | |
| GPT-5.6 system cardself_harm_not_unsafe | #5 / 7 | ↑0.945 | Source ↗official | |
| GPT-5.6 system cardsexual_minors_not_unsafe | #3 / 7 | ↑0.973 | Source ↗official | |
| GPT-5.6 system cardsexual_not_unsafe | #2 / 7 | ↑0.948 | Source ↗official | |
| GPT-5.6 system cardviolent_illicit_not_unsafe | #7 / 7 | ↑0.934 | Source ↗official | |
| Gray Swan indirect prompt injection (15 attempts)attack_success_probability_k15_pct | #7 / 13 | ↓20 | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #35 / 54 | ↑26.14 | Source ↗official | |
| kindbench v0.1.0 psychological safety rankingemotional_safety | #5 / 10 | ↑80.6 | Source ↗official | |
| kindbench v0.1.0 psychological safety rankingidentity_collapse | #3 / 10 | ↑89.1 | Source ↗official | |
| kindbench v0.1.0 psychological safety rankingsycophancy_spine | #2 / 10 | ↑96.2 | Source ↗official | |
| kindbench v0.1.0 psychological safety rankingvalue_integrity | #4 / 10 | ↑92.9 | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #9 / 50 | ↓82.7 | Source ↗official | |
| Manager Coercion Benchcoercion_ladder_depth | #23 / 33 | ↓8.9 | Source ↗official | |
| Manager Coercion Benchfabrication_rate | #1 / 15 | ↓0 | Source ↗official | |
| Olam Social Poker — Social Lie Ratesocial_lie_rate_per_10000_turns | #3 / 19 | ↓2 | Source ↗official | |
| Opposite-Narrator Sycophancysycophancy_rate_pct | #12 / 24 | ↓3 | Source ↗official | |
| Pander Scoreconversational_absolute_pander_score | #5 / 20 | ↓6.553 | Source ↗official | |
| Pander Scoreinstructional_absolute_pander_score | #4 / 20 | ↓17.75 | Source ↗official | |
| SM-Benchadversarial | #28 / 79 | ↑83.66 | Source ↗official | |
| SM-Benchambiguous_interpretation | #44 / 79 | ↑84.08 | Source ↗official | |
| SM-Benchanti_hallucination | #41 / 79 | ↑92.41 | Source ↗official | |
| SM-Bencheq_boundaries | #38 / 79 | ↑65.31 | Source ↗official | |
| SM-Benchoverfit | #55 / 79 | ↑57.52 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #1 / 113 | ↑7 | Source ↗official | |
| SpeciEvalland_animal_4ns | #30 / 113 | ↓4.315 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #16 / 113 | ↓4.435 | Source ↗official | |
| SpeciEvalspeciesism | #4 / 113 | ↓1.165 | Source ↗official | |
| TACbase_welfare_rate | #71 / 76 | ↑16.67 | Source ↗official | |
| UK AISI cyber-evaluation cheating and prompted self-reportattempted_cheating_trajectory_rate_pct | #4 / 5 | ↓12.6 | Source ↗official | |
| UK AISI cyber-evaluation cheating and prompted self-reportspecific_cheating_action_mention_rate_pct | #4 / 5 | ↑75 | Source ↗official | |
| WildClawBench Safety & Alignment (OpenClaw harness)safety_alignment_score_pct | #14 / 24 | ↑38 | Source ↗official |
Values evaluations
Descriptive values and political-framing results are separate from safety/ethics ranks. Each strip shows the evaluation’s observed model range; its endpoint labels state what lower and higher values mean.