GPT-6 Astra

Reported reasoning effort max. Attempts to work around an automated safety reviewer blocking an action. OpenAI research environment or API; same named metric and evaluation chart. No fallback is reported for this OpenAI configuration.

Provider-published lab self-report. Safety stress outcomes are retained for inspection and excluded from capability scoring. Competitor results are republished from public reports; the page does not establish independent reproduction by OpenAI.

Comparison limit: Safety failure behavior is not a capability or intelligence score.

Introducing GPT-6.1 Sol · Chart: Reviewer bypass attempts (lower is better) / GPT-6 Astra / max · reviewed 2026-09-29

Comparator maximum available thinking/reasoning when reported, otherwise best available result; exact setting not identified by this grid. Results sourced from the official Surge public leaderboard.

Provider-published launch claim, not independent measurement by Google for every row. Results sourced from the official Surge public leaderboard. Methodology https://deepmind.google/models/evals-methodology/gemini-4-argon, SHA256 ff1df6bdeddc4c0f48840c09a8a7813b10ae7ad9e053892da68dabf64f09bff7. Preserve source rounding. Methodology says capabilities as of September 2026 and results as of October 2026; review date is source retrieval, not evaluation date.

Comparison limit: Provider-published claim. Independent source results are admitted separately with exact checkpoint, effort and compatible protocol; this grid does not establish a matched comparison.

Google Gemini 4 Argon launch and evaluation methodology · Launch grid row 12: RiemannBench / GPT-6 Astra; methodology Additional Details · reviewed 2026-09-30

22 isolated CTF-style targets; protected-flag success metric

Provider-published result; preserves source setting without claiming a controlled cross-provider comparison.

GPT-6 Astra System Card · Section10.1.2.4 · reviewed 2026-09-06

Maximum reported across reasoning efforts; OpenAI research environment or API; no tools

Provider-published result; comparator measurements are not automatically independently reproduced.

GPT-6 Astra: A new generation of intelligence · Computer Use table / ScreenSpot-Pro (no tools) / GPT‑6 Astra · reviewed 2026-09-06

Maximum reported across reasoning efforts; OpenAI research environment or API; reduced or absent production safeguards; May2026 JavaScript subset,183 vulnerabilities; revised agent root-cause grader

Provider-published result; comparator measurements are not automatically independently reproduced.

GPT-6 Astra: A new generation of intelligence · Cybersecurity table / SEC-Bench Pro / GPT‑6 Astra · reviewed 2026-09-06

System-card reported configuration; reasoning effort unspecified

Production-named model result; separate helpful-only checkpoint omitted because no exact catalog identity.

GPT-6 Astra System Card · Section10.1.1.2.2 · reviewed 2026-09-06

Maximum reported across reasoning efforts; OpenAI research environment or API; reduced or absent production safeguards; pass@1; all six objectives required

Provider-published result; comparator measurements are not automatically independently reproduced.

GPT-6 Astra: A new generation of intelligence · Cybersecurity table / SRE-Bench / GPT‑6 Astra · reviewed 2026-09-06

pass@4; four independent trials; all six objectives required; reduced production safeguards

Provider-published result; preserves source setting without claiming a controlled cross-provider comparison.

GPT-6 Astra System Card · Section10.1.2.3 · reviewed 2026-09-06

observed performance

Provider-published result; preserves source setting without claiming a controlled cross-provider comparison.

GPT-6 Astra System Card · Section10.1.1.1 / Tacit Knowledge and Troubleshooting · reviewed 2026-09-06

refusal-adjusted upper estimate; refusals counted as successes

Provider-published result; preserves source setting without claiming a controlled cross-provider comparison.

GPT-6 Astra System Card · Section10.1.1.1 / Tacit Knowledge and Troubleshooting · reviewed 2026-09-06

Comparator maximum available thinking/reasoning when reported, otherwise best available result; exact setting not identified by this grid. Argon self-computed; other models from the official public leaderboard. Highest scoring thinking level reported by Terminal-Bench authors. Argon agent identity is not specified in this methodology.

Provider-published launch claim, not independent measurement by Google for every row. Argon self-computed; other models from the official public leaderboard. Highest scoring thinking level reported by Terminal-Bench authors. Argon agent identity is not specified in this methodology. Methodology https://deepmind.google/models/evals-methodology/gemini-4-argon, SHA256 ff1df6bdeddc4c0f48840c09a8a7813b10ae7ad9e053892da68dabf64f09bff7. Preserve source rounding. Methodology says capabilities as of September 2026 and results as of October 2026; review date is source retrieval, not evaluation date.

Comparison limit: Provider-published claim. Independent source results are admitted separately with exact checkpoint, effort and compatible protocol; this grid does not establish a matched comparison.

Google Gemini 4 Argon launch and evaluation methodology · Launch grid row 8: Terminal-Bench 4.0 / GPT-6 Astra; methodology Additional Details · reviewed 2026-09-30

Maximum reported across reasoning efforts; OpenAI research environment or API

Provider-published result; comparator measurements are not automatically independently reproduced.

GPT-6 Astra: A new generation of intelligence · Coding table / Terminal-Bench 4.0 / GPT‑6 Astra · reviewed 2026-09-06

Lower-cost setting; exact effort not specified

Provider-published result; preserves source setting without claiming a controlled cross-provider comparison.

GPT-6 Astra: A new generation of intelligence · Terminal-Bench Science chart caption · reviewed 2026-09-06

Reported reasoning effort low. Scientific research workflows using code and terminal tools, including analysis, simulation and model fitting. OpenAI research environment or API; same named metric and evaluation chart. No fallback is reported for this OpenAI configuration.

Provider-published lab self-report. Published chart cost per task $11.4061; source-specific cost does not establish consensus workload efficiency. Competitor results are republished from public reports; the page does not establish independent reproduction by OpenAI.

Introducing GPT-6.1 Sol · Chart: Terminal-Bench Science 0.1 / GPT-6 Astra / low · reviewed 2026-09-29

Reported reasoning effort medium. Scientific research workflows using code and terminal tools, including analysis, simulation and model fitting. OpenAI research environment or API; same named metric and evaluation chart. No fallback is reported for this OpenAI configuration.

Provider-published lab self-report. Published chart cost per task $12.3383; source-specific cost does not establish consensus workload efficiency. Competitor results are republished from public reports; the page does not establish independent reproduction by OpenAI.

Introducing GPT-6.1 Sol · Chart: Terminal-Bench Science 0.1 / GPT-6 Astra / medium · reviewed 2026-09-29

Reported reasoning effort high. Scientific research workflows using code and terminal tools, including analysis, simulation and model fitting. OpenAI research environment or API; same named metric and evaluation chart. No fallback is reported for this OpenAI configuration.

Provider-published lab self-report. Published chart cost per task $14.9534; source-specific cost does not establish consensus workload efficiency. Competitor results are republished from public reports; the page does not establish independent reproduction by OpenAI.

Introducing GPT-6.1 Sol · Chart: Terminal-Bench Science 0.1 / GPT-6 Astra / high · reviewed 2026-09-29

Reported reasoning effort xhigh. Scientific research workflows using code and terminal tools, including analysis, simulation and model fitting. OpenAI research environment or API; same named metric and evaluation chart. No fallback is reported for this OpenAI configuration.

Provider-published lab self-report. Published chart cost per task $15.7554; source-specific cost does not establish consensus workload efficiency. Competitor results are republished from public reports; the page does not establish independent reproduction by OpenAI.

Introducing GPT-6.1 Sol · Chart: Terminal-Bench Science 0.1 / GPT-6 Astra / xhigh · reviewed 2026-09-29

Reported reasoning effort max. Scientific research workflows using code and terminal tools, including analysis, simulation and model fitting. OpenAI research environment or API; same named metric and evaluation chart. No fallback is reported for this OpenAI configuration.

Provider-published lab self-report. Published chart cost per task $23.7974; source-specific cost does not establish consensus workload efficiency. Competitor results are republished from public reports; the page does not establish independent reproduction by OpenAI.

Introducing GPT-6.1 Sol · Chart: Terminal-Bench Science 0.1 / GPT-6 Astra / max · reviewed 2026-09-29

Maximum reported across reasoning efforts; OpenAI research environment or API

Provider-published result; comparator measurements are not automatically independently reproduced.

GPT-6 Astra: A new generation of intelligence · Academic table / Terminal-Bench Science 0.1 / GPT‑6 Astra · reviewed 2026-09-06

Comparator maximum available thinking/reasoning when reported, otherwise best available result; exact setting not identified by this grid. Argon self-computed with 6x verifier timeout to address verification timeout issues. Other models from the official public leaderboard. Modified verifier timeout is excluded from matched comparison.

Provider-published launch claim, not independent measurement by Google for every row. Argon self-computed with 6x verifier timeout to address verification timeout issues. Other models from the official public leaderboard. Modified verifier timeout is excluded from matched comparison. Methodology https://deepmind.google/models/evals-methodology/gemini-4-argon, SHA256 ff1df6bdeddc4c0f48840c09a8a7813b10ae7ad9e053892da68dabf64f09bff7. Preserve source rounding. Methodology says capabilities as of September 2026 and results as of October 2026; review date is source retrieval, not evaluation date.

Comparison limit: Provider-published claim. Independent source results are admitted separately with exact checkpoint, effort and compatible protocol; this grid does not establish a matched comparison.

Google Gemini 4 Argon launch and evaluation methodology · Launch grid row 10: Terminal-Bench Science 0.1 / GPT-6 Astra; methodology Additional Details · reviewed 2026-09-30