Claude Opus 5.5 · FrontierCode 2.0
Coding · provider-reported by Anthropic. Evaluation snapshot: unknown. Test date: unknown.
Effort: Highest available (source wording; numeric budget unknown). Safeguards on. Agent/harness, token budget, attempts/pass@k and test date not disclosed in the announcement.
Exact public release label matched to catalog ID; evaluation snapshot not disclosed. Not a guarantee for future revisions of the API alias.
Original report · Methodology / system card · Source checked 2026-11-12 (not the test date).
Claude Opus 5.5 · Humanity’s Last Exam (with tools; version unspecified)
Reasoning · provider-reported by Anthropic. Evaluation snapshot: unknown. Test date: unknown.
Effort: Highest available (source wording; numeric budget unknown). Tools enabled. Safeguards on. Agent/harness, token budget, attempts/pass@k and test date not disclosed in the announcement.
Exact public release label matched to catalog ID; evaluation snapshot not disclosed. Not a guarantee for future revisions of the API alias.
Original report · Methodology / system card · Source checked 2026-11-12 (not the test date).
Claude Opus 5.5 · Chartography (no tools; version unspecified)
Visual reasoning · provider-reported by Anthropic. Evaluation snapshot: unknown. Test date: unknown.
Effort: Highest available (source wording; numeric budget unknown). No tools. Safeguards on. Agent/harness, token budget, attempts/pass@k and test date not disclosed in the announcement.
Exact public release label matched to catalog ID; evaluation snapshot not disclosed. Not a guarantee for future revisions of the API alias.
Original report · Methodology / system card · Source checked 2026-11-12 (not the test date).
Gemini 4 Hydrogen · SWE-sweep (version unspecified)
Coding · provider-reported by Google DeepMind. Evaluation snapshot: unknown. Test date: unknown. Reported period: November 2026 (not an exact test date).
Gemini API, highest thinking; Antigravity-CLI 1.4.8; safety filters on; subtask-weighted reward; single attempt/pass@1. Highest thinking is source wording, not an inferred token budget.
Exact public release label matched to catalog ID; evaluation snapshot not disclosed. Not a guarantee for future revisions of the API alias.
Original report · Methodology / system card · Source checked 2026-11-12 (not the test date).
Gemini 4 Hydrogen · Terminal-Bench 4.1 (cheating-adjusted)
Coding · provider-reported by Google DeepMind. Evaluation snapshot: unknown. Test date: unknown. Reported period: November 2026 (not an exact test date).
Gemini API, high effort; mini-swe-agent vision harness; average reward over 5 attempts per task (not best-of-5). Highest thinking is source wording, not an inferred token budget.
Exact public release label matched to catalog ID; evaluation snapshot not disclosed. Not a guarantee for future revisions of the API alias.
Original report · Methodology / system card · Source checked 2026-11-12 (not the test date).
Gemini 4 Hydrogen · Agents’ Last Exam v1.1
Knowledge work · provider-reported by Google DeepMind. Evaluation snapshot: unknown. Test date: unknown. Reported period: November 2026 (not an exact test date).
Gemini API, highest thinking; Antigravity-CLI 1.4.8; 122 tasks; 5-hour window; Oct 17 task version, not Oct 26 update; safety filters on; averaged rubric score; pass@1. Highest thinking is source wording, not an inferred token budget.
Exact public release label matched to catalog ID; evaluation snapshot not disclosed. Not a guarantee for future revisions of the API alias.
Original report · Methodology / system card · Source checked 2026-11-12 (not the test date).
Gemini 4 Hydrogen · OSWorld 2.1 (82-task offline subset)
Computer use · provider-reported by Google DeepMind. Evaluation snapshot: unknown. Test date: unknown. Reported period: November 2026 (not an exact test date).
Highest thinking; best of 3 runs, single attempt per run; partial score; official Docker/evaluator; 1080p; max 500 steps; updated Gemini CUA harness, parallel batch tools, full screenshot history with compaction; PyAutoGUI; safety filters on. Highest thinking is source wording, not an inferred token budget.
Exact public release label matched to catalog ID; evaluation snapshot not disclosed. Not a guarantee for future revisions of the API alias.
Original report · Methodology / system card · Source checked 2026-11-12 (not the test date).
Gemini 4 Hydrogen · Chartography (no tools; version unspecified)
Visual reasoning · provider-reported by Google DeepMind. Evaluation snapshot: unknown. Test date: unknown. Reported period: November 2026 (not an exact test date).
Gemini API, highest thinking; no tools; self-computed following Surge public leaderboard; pass@1. Exact model snapshot and benchmark version not disclosed. Highest thinking is source wording, not an inferred token budget.
Exact public release label matched to catalog ID; evaluation snapshot not disclosed. Not a guarantee for future revisions of the API alias.
Original report · Methodology / system card · Source checked 2026-11-12 (not the test date).