Ranks compare reported scores in the same published row; ties share a rank.
Terminal-Bench 4.0. Claude Code; Haiku safeguards enabled without fallback. GPT-6 Luna is quoted from the public Codex leaderboard.
All eight results and four comparison columns from Anthropic’s announcement, read 8 October 2026. Haiku 5.5 uses max effort; Sonnet’s FrontierCode uses xhigh. HLE is split into no-tools and with-tools rows. OSWorld uses partial credit on the official 82-task offline subset, differing from earlier combined-subset reports. AA ran GDPval-AA and AA-Briefcase; these rounded quotes from Anthropic remain separate from the directly imported public AA measurements. GPT results retain the source and configuration described in the system card. Missing cells stay blank; the reported Haiku 4.5 Terminal-Bench score is zero. Each rank compares one published row only.
www.anthropic.com/claude-haiku-5-5
Ranks compare reported scores in the same published row; ties share a rank.
Terminal-Bench 4.0. §8.4: Haiku 10 trials per task, SE ±1.9 pts; 1.8% flagged trials failed without fallback. Sonnet/Opus use five trials and fallback; Opus xhigh. Luna/Sol quote public Codex max-effort results.
Section 8, pages 111–139 of the Haiku 5.5 system card, read 8 October 2026. Values come from exact prose, tables and labelled figures. Unless specified, Haiku uses adaptive thinking at max effort and five trials with evaluation-specific context up to 1M. Each row retains its subset, metric, effort and source notes; no score is assigned to an unspecified SWE-Bench Pro version. Terminal runs use safeguards without fallback and no internet egress. ProgramBench uses filtered tasks and hidden-test pass rate without a time limit. OSWorld partial and strict results use the official 82-task offline subset. CAD uses 0–1 voxel IoU. Healthcare raw and length-adjusted scores stay separate. Life-science results disable Haiku biology safeguards. Independently run figures quoted by Anthropic remain publisher reports here. Missing comparison values are not inferred from rounded charts.
www.anthropic.com/claude-haiku-5-5-system-card
Ranks compare reported scores in the same published row; ties share a rank.
Terminal-Bench 4.0. Claude Code harness. The September V4.1 technical-report snapshot used a different harness; retain this release comparison separately.
The complete 11-row, six-model comparison from DeepSeek’s dated MIT model card, read 17 October 2026 at revision 1bae0e1088993b9b82de08f59de5be5a34e42170. All 63 published comparison cells and both additional parenthesized HLE scores are retained as 65 figures across 12 rows. The HLE text-only scores are split into their own row and benchmark page; GLM-5.3’s dagger-marked 42.0 is text-only and has no full-set score. Unreported values stay null and do not enter ranks. K3 and Opus 5 are displayed as Kimi K3 and Claude Opus 5. Terminal-Bench 4.0 uses Claude Code, DeepSWE v1.1 uses Mini-SWE-V2, and the release notes identify ZeroBench-main as Pass@5. Other effort levels, trial counts and judges are not fully specified. These are DeepSeek publication claims, separate from independent evaluations and earlier publisher snapshots.
huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017
Ranks compare reported scores in the same published row; ties share a rank.
Page 5 of Google DeepMind’s model evaluation, read 1 October 2026. Gemini runs use the API at the highest thinking setting, generally pass@1 without majority voting or parallel test-time compute; smaller benchmarks average multiple trials. Other models use their highest available published reasoning results. Google quotes Vals AI for knowledge-work and Vibe Code results, Zapier’s private-set AutomationBench leaderboard, Proximal for FrontierSWE v2, Surge AI for RiemannBench and no-tools Chartography, and the CWE-bench v1 leaderboard. Argon’s DeepSWE uses mini-swe; Terminal-Bench 4.0 is self-computed for Argon and quoted from the author board for other models. PostTrainBench v1.1 uses OpenCode with one H100 and a 10-hour budget for all models. Terminal-Bench Science uses a 6x verifier timeout for Argon. LABBench 2 uses terminal, bioinformatics tools, Python, R and internet access for all models. GraphWalks uses BFS F1 on two separate subsets: 650 items up to 128k tokens and 200 items from 256k to 1M. Agent’s Last Exam uses ALE-Claw, a 5-hour window and binary pass rate, with safety filters enabled for Argon. OSWorld 2.0 uses offline partial reward, the best of three single-attempt runs for Argon, safety filters, 1080p, 500 steps, Gemini CUA with parallel batch tools and compaction, screenshot-only observation and the 08.08 patch; Claude’s combined online/offline results are omitted. LVBench runs without tools, using 1 FPS for Gemini, 800 frames for Astra, 300 for Fable and 600 for Opus. Missing cells remain blank. These are the report’s published figures, preserved separately from this site’s independent source snapshots.
storage.googleapis.com/deepmind-media/gemini/gemini_4_argon_model_evaluation.pdf
Ranks compare reported scores in the same published row; ties share a rank.
Table 3 of the V4.1 technical report, published with the model release. Every column is the vendor's own run at its maximum reasoning effort, each on that vendor's own harness, so this compares publication claims rather than one shared run. DeepSeek renamed nothing and disclosed no fallback rates; the report instead documents the opposite problem, agents searching for ways to game the harness, and says its visual-agent rows were run with a 512k-token context. The benchmark names are reproduced as DeepSeek wrote them — including rows this site keeps apart, such as HLE with and without tools — and the model columns are renamed to the names used on this site. DeepSeek's table is also republished, unchanged, on the model page at ollama.com; that copy prints 64.0 for the V4.1 row of NL2Repo-Bench where the report prints 65.4, and this site follows the report.
huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/resolve/main/DeepSeek_V41_Tech_Report.pdf
Ranks compare reported scores in the same published row; ties share a rank.
OpenAI reports the maximum score at any effort level, in its own research environment or through its API. Claude and Gemini figures are taken by OpenAI from those vendors' published reports, so this table is a comparison between publication claims, not between runs.
openai.com/index/gpt-6-astra/
Ranks compare reported scores in the same published row; ties share a rank.
Anthropic runs its own harness and states that margin size is a weakening guide to real differences at this level. Two rows rely on other publishers: AutomationBench figures come from Zapier, and the GPT figures are as reported by OpenAI. The Claude Opus 5.5 row was run with production safeguards enabled, so some security, biology and frontier-development tasks were completed by a fallback model.
www.anthropic.com/news/claude-opus-5-5
Ranks compare reported scores in the same published row; ties share a rank.
Terminal-Bench 4.0. Anthropic's own harness. Vals, running the same benchmark independently, scores Sonnet 5.5 at 53.03% and Opus 5.5 at 61.62%, so this row and the Vals board above are two different runs of one benchmark. Opus 5.5 is reported at Xhigh effort, its highest score.
Anthropic publishes this table with Sonnet 5.5, the second model in its Claude 5.5 line, and states that Sonnet 5.5 at Max effort performs comparably to Opus 5.5 on several evaluations while Opus 5.5 remains stronger on open-ended work. The GPT-6 Sol column is quoted from OpenAI, not re-run, and Anthropic notes an OpenAI image-understanding bug that may not yet be reflected in third-party scores. Two rows come from other publishers: Artificial Analysis ran GDPval-AA and AA-Briefcase on a pre-release Sonnet 5.5 deployment that Anthropic later found had a structured-output bug, and Surge AI runs Chartography. Sonnet 5.5 launched with the cyber safeguards first built for Opus-class models, so some security work was completed by a fallback model.
www.anthropic.com/claude-sonnet-5-5