GPT-6.1 Sol: score and benchmark evidence
Historical release · · OpenAI
125.4 points · Rank 11 among 39 ranked models · Stability range 114.4–136.8 points.
Rank follows the point estimate; it does not establish statistically significant superiority. The scale is anchored at mean 100, SD 15 in the frozen calibration cohort, not human IQ or a percentage.
One score summarizes demonstrated capability across published reasoning settings and seven equally weighted domains. Related benchmark variants share a family budget, and influence from one evaluator is capped.
Release tli-2026-v1.0-2026-10-28-gemini-4-carbon-lab-11. Methodology and limitations for this release.
This page preserves historical evidence. See the current model record.
Capability domains
Domain scores use calibration-cohort standard-deviation units. Unmeasured domains are shown as unavailable, not zero.
| Domain | Score (SD units) | Observed benchmarks |
|---|---|---|
| Knowledge & reasoning | 1.00 | 5 |
| Coding & software engineering | 0.78 | 3 |
| Agentic tool & computer use | 0.92 | 2 |
| Professional & real-world work | 0.42 | 7 |
| Multimodal & vision | 0.51 | 2 |
| Cybersecurity | Unavailable | 0 |
| Preference & communication | Unavailable | 0 |
Published benchmark evidence
19 results contribute to this score, from 24 collected results and 19 contributing benchmark families. Counts are not independent sample sizes.
Results, source citations, review decisions and compatible reasoning modes are stored in a versioned relational database. Each publication uses a sealed, checksum-verified export; CSV files are optional exports, not editable scoring inputs. Previous revisions remain available for audit. Collected means a numeric result exists in the database. Used means its source, exact model checkpoint and benchmark protocol have been reviewed and its benchmark has a frozen calibration supported by multiple providers. For each benchmark we select the highest published aggregate across comparable reasoning modes, not necessarily maximum effort, and count it once. More tested settings provide more chances to record a high score; this selection advantage is not corrected by the conditional interval. Developer-reported and sole/unspecified settings remain eligible and disclosed. Unresolved does not mean false or useless: conflicting metrics, unverified citations and unmatched versions stay collected until reconciled.
The score stability range is a 95% bootstrap score interval conditional on frozen calibration and comparable evidence availability. It resamples benchmark families and evaluators and adds a frozen missing-evidence correction. It excludes selection across reasoning settings, selective publication, calibration changes and new domains. It is not a 95% guarantee about true capability or where a future score will land. Sparse or conflicting evidence can widen the range.
Reporting sensitivity: ±11.7 points. Sensitivity to selectively published results. This stress range is not a confidence interval or a probability statement.
AA-Briefcase v1.1 (Elo): 1555.1 native-points
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (max); Artificial Analysis independent evaluation, 2026-10-27; fallback not reported
Metric: published native-points
Original source · Retrieved 2026-10-27
Daily evaluator refresh; replaces 1556.34 recorded 2026-10-21.
GDPval-AA v2.1 (Elo): 1607.61 native-points
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (max); Artificial Analysis independent evaluation, 2026-10-27; fallback not reported
Metric: published native-points
Original source · Retrieved 2026-10-27
Daily evaluator refresh; replaces 1592.1 recorded 2026-10-21.
AutomationBench-AA: 66.5737214136%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (xhigh); Artificial Analysis independent evaluation, September 30 2026; fallback not reported
Metric: published percent
Original source · Retrieved 2026-09-30
Exact model and effort in the evaluator chart payload. HLE is text-only; Omniscience is the native index, not accuracy; AutomationBench-AA is guardrail-adjusted objective credit, not workflow pass rate.
Terminal-Bench 4.0 — AA harness: 56.0606060606%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (max); Artificial Analysis independent evaluation, September 30 2026; fallback not reported
Metric: published percent
Original source · Retrieved 2026-09-30
Exact model and effort in the evaluator chart payload. HLE is text-only; Omniscience is the native index, not accuracy; AutomationBench-AA is guardrail-adjusted objective credit, not workflow pass rate.
SciCode: 55.787037037%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (high); Artificial Analysis independent evaluation, September 30 2026; fallback not reported
Metric: published percent
Original source · Retrieved 2026-09-30
Exact model and effort in the evaluator chart payload. HLE is text-only; Omniscience is the native index, not accuracy; AutomationBench-AA is guardrail-adjusted objective credit, not workflow pass rate.
HLE text-only — AA harness: 52.9193697868%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (max); Artificial Analysis independent evaluation, September 30 2026; fallback not reported
Metric: published percent
Original source · Retrieved 2026-09-30
Exact model and effort in the evaluator chart payload. HLE is text-only; Omniscience is the native index, not accuracy; AutomationBench-AA is guardrail-adjusted objective credit, not workflow pass rate.
GDP.pdf — AA harness: 32%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (high); Artificial Analysis independent evaluation, September 30 2026; fallback not reported
Metric: published percent
Original source · Retrieved 2026-09-30
Exact model and effort in the evaluator chart payload. HLE is text-only; Omniscience is the native index, not accuracy; AutomationBench-AA is guardrail-adjusted objective credit, not workflow pass rate.
CritPt: 31.7142857143%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (xhigh); Artificial Analysis independent evaluation, September 30 2026; fallback not reported
Metric: published percent
Original source · Retrieved 2026-09-30
Exact model and effort in the evaluator chart payload. HLE is text-only; Omniscience is the native index, not accuracy; AutomationBench-AA is guardrail-adjusted objective credit, not workflow pass rate.
AA-Omniscience: 41.5333333333 native-points
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (max); Artificial Analysis independent evaluation, September 30 2026; fallback not reported
Metric: published native-points
Original source · Retrieved 2026-09-30
Restored on 2026-10-28: 42.033333 recorded 2026-10-28 came from the evaluator's renamed AA-Omniscience (AA-Omniscience v1.1: Knowledge and Hallucination Benchmark), a different version, so this row keeps its earlier value. Exact model and effort in the evaluator chart payload. HLE is text-only; Omniscience is the native index, not accuracy; AutomationBench-AA is guardrail-adjusted objective credit, not workflow pass rate.
AA-LCR v1.1: 84%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (low); Artificial Analysis independent evaluation, September 30 2026; fallback not reported
Metric: published percent
Original source · Retrieved 2026-09-30
Exact model and effort in the evaluator chart payload. HLE is text-only; Omniscience is the native index, not accuracy; AutomationBench-AA is guardrail-adjusted objective credit, not workflow pass rate.
MMMU-Pro: 85.9537572254%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (max); Artificial Analysis independent evaluation, September 30 2026; fallback not reported
Metric: published percent
Original source · Retrieved 2026-09-30
Independent MMMU-Pro aggregate for this effort.
AA Intelligence Index v4.3.2 (score): 51.8332597012 native-points
Not used in the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (max); Artificial Analysis independent evaluation, September 30 2026; fallback not reported
Metric: published native-points
Original source · Retrieved 2026-09-30
Restored on 2026-10-28: 54.508734 recorded 2026-10-28 came from the evaluator's renamed AA Intelligence Index v4.3.2 (score) (Artificial Analysis Intelligence Index v5.0), a different version, so this row keeps its earlier value. Reference composite only, never scored alongside its component evaluations.
HealthBench Professional — Sep 22 length-adjusted: 64.2%
Not used in the score. Developer-reported. Review status: reviewed-developer.
Configuration: GPT-6.1 Sol (undisclosed); OpenAI September 30 system-card addendum; fallback not reported
Metric: published percent
Original source · Retrieved 2026-09-30
Table 7 length-adjusted column, not the parenthesised unadjusted score. Same length-adjusted protocol as the September 22 release rows; remains uncalibrated.
HealthBench Overall — Sep 22 length-adjusted: 58.5%
Not used in the score. Developer-reported. Review status: reviewed-developer.
Configuration: GPT-6.1 Sol (undisclosed); OpenAI September 30 system-card addendum; fallback not reported
Metric: published percent
Original source · Retrieved 2026-09-30
Table 7 length-adjusted column, not the parenthesised unadjusted score. Same length-adjusted protocol as the September 22 release rows; remains uncalibrated.
HealthBench Hard — Sep 22 length-adjusted: 36.2%
Not used in the score. Developer-reported. Review status: reviewed-developer.
Configuration: GPT-6.1 Sol (undisclosed); OpenAI September 30 system-card addendum; fallback not reported
Metric: published percent
Original source · Retrieved 2026-09-30
Table 7 length-adjusted column, not the parenthesised unadjusted score. Same length-adjusted protocol as the September 22 release rows; remains uncalibrated.
HealthBench Consensus — Sep 22 length-adjusted: 96%
Not used in the score. Developer-reported. Review status: reviewed-developer.
Configuration: GPT-6.1 Sol (undisclosed); OpenAI September 30 system-card addendum; fallback not reported
Metric: published percent
Original source · Retrieved 2026-09-30
Table 7 length-adjusted column, not the parenthesised unadjusted score. Same length-adjusted protocol as the September 22 release rows; remains uncalibrated.
BioMysteryBench v1 — Vals: 79.63%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (max); Vals AI BioMysteryBench harness; provider OpenAI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-21
Vals AI model page, headline accuracy 79.63 ± 0.74, rank 2/27. Vals notes some benchmarks may use a different provider or parameters.
Legal Research Bench — Vals: 38.46%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (max); Vals AI Legal Research Bench harness; provider OpenAI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-21
Vals AI model page, headline accuracy 38.46 ± 3.38, rank 37/82. Vals notes some benchmarks may use a different provider or parameters.
MedCode — Vals: 48.84%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (max); Vals AI MedCode harness; provider OpenAI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-21
Vals AI model page, headline accuracy 48.84 ± 2.12, rank 28/107. Vals notes some benchmarks may use a different provider or parameters.
MedScribe — Vals: 86.45%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (max); Vals AI MedScribe harness; provider OpenAI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-21
Vals AI model page, headline accuracy 86.45 ± 1.9, rank 23/109. Vals notes some benchmarks may use a different provider or parameters.
ProofBench v1.1 — Vals: 99%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (max); Vals AI ProofBench v1.1 harness; provider OpenAI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-21
Vals AI model page, headline accuracy 99.0 ± 1.0, rank 4/50. Vals notes some benchmarks may use a different provider or parameters.
Public Benefits Bench — Vals: 59.34%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (max); Vals AI Public Benefits Bench v1.1 harness; provider OpenAI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-21
Vals AI model page, headline accuracy 59.34 ± 1.28, rank 35/50. Vals notes some benchmarks may use a different provider or parameters.
SAGE — Vals: 46.54%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (max); Vals AI SAGE harness; provider OpenAI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-21
Vals AI model page, headline accuracy 46.54 ± 3.46, rank 35/91. Vals notes some benchmarks may use a different provider or parameters.
Vibe Code Bench v1.1 — Vals: 88.93%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: GPT-6.1 Sol (max); Vals AI Vibe Code Bench v1.1 harness; provider OpenAI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-21
Vals AI model page, headline accuracy 88.93 ± 1.89, rank 8/116. Vals notes some benchmarks may use a different provider or parameters.
Cite this model record
The Latent. “GPT-6.1 Sol: score and benchmark evidence.” 2026-10-28. Release tli-2026-v1.0-2026-10-28-gemini-4-carbon-lab-11.
Download release data and definitions (JSON) · Data usage terms