gemini-4-argon-high
LM Arena overall rank: 1; rating: 1525.22; votes: 4,932.
Organization: google; published October 2, 2026.
Model VS evidence index
Compare AI models with separate, traceable evidence
Model VS keeps user preference, objective benchmark results, and specialist task data separate. It does not create a synthetic score or declare an overall winner.
Sources and current snapshots
LM Arena dataset lmarena-ai/leaderboard-dataset was synced October 6, 2026; leaderboard published October 2, 2026. LiveBench release 2026-06-25 was synced October 6, 2026. Specialist data comes from Hugging Face official leaderboard API, synced October 6, 2026.
Method: Source records are shown in their native measurement systems. Rank, user-preference rating, benchmark scores, and task-specific percentages are not converted into one score.
Limits: Benchmark measurements are snapshots, not guarantees of real-world behavior. They can change with source updates and should be considered with task requirements, pricing, safety, and licensing.
Current LM Arena records
claude-opus-4-6-high
LM Arena overall rank: 2; rating: 1504.73; votes: 77,636.
Organization: anthropic; published October 2, 2026.
claude-fable-5-high
LM Arena overall rank: 3; rating: 1504.27; votes: 38,387.
Organization: anthropic; published October 2, 2026.
claude-opus-5.5-high
LM Arena overall rank: 4; rating: 1503.70; votes: 4,552.
Organization: anthropic; published October 2, 2026.
Current LiveBench records
claude-opus-4-5-20251101-thinking-64k-high-effort
LiveBench raw metrics: AMPS Hard 99.0; code completion 80.435; code generation 78.873.
Release: 2026-06-25; synced October 6, 2026.
claude-opus-4-6-thinking-auto-high-effort
LiveBench raw metrics: AMPS Hard 97.0; code completion 76.087; code generation 80.282.
Release: 2026-06-25; synced October 6, 2026.
claude-opus-4-7-xhigh-effort
LiveBench raw metrics: AMPS Hard 98.0; code completion 78.261; code generation 85.915.
Release: 2026-06-25; synced October 6, 2026.
claude-sonnet-4-6-thinking-auto-medium-effort
LiveBench raw metrics: AMPS Hard 76.0; code completion 78.261; code generation 80.282.
Release: 2026-06-25; synced October 6, 2026.
Current specialist benchmark records
ornith-ai/Ornith-1.5-397B
SWE-bench Verified: rank 1; % resolved 86.
Community submission — not marked verified. Dataset: SWE-bench/SWE-bench_Verified; source: Ornith-1.5-397B model card; benchmark data updated August 16, 2026.
mindlab-research/Macaron-V1-Venti
SWE-bench Verified: rank 2; % resolved 85.6.
Community submission — not marked verified. Dataset: SWE-bench/SWE-bench_Verified; source: Model Card; benchmark data updated August 16, 2026.
mindlab-research/Macaron-V1-Coding-Venti
SWE-bench Verified: rank 3; % resolved 85.6.
Community submission — not marked verified. Dataset: SWE-bench/SWE-bench_Verified; source: Model Card; benchmark data updated August 16, 2026.
ornith-ai/Ornith-1.0-397B
SWE-bench Verified: rank 4; % resolved 82.4.
Community submission — not marked verified. Dataset: SWE-bench/SWE-bench_Verified; source: Ornith-1.0-397B model card; benchmark data updated August 16, 2026.
Open the interactive comparison to filter sources and inspect the complete evidence record.
Turn a score into a decision
Model VS calculates these differences from published snapshots. The measurements belong to the named sources; the interpretation is ours.
Read preference ratings with their uncertainty
These are the first two ranked records in the saved Arena snapshot, published 2026-10-02 and retrieved 2026-10-06T09:00:53.013Z. Model VS rounds the display to two decimals; the linked JSON retains source precision.
gemini-4-argon-high
Source rank: 1. Rating: 1525.22.
Source 95% interval: 1516.39–1534.04.
Votes: 4932.
claude-opus-4-6-high
Source rank: 2. Rating: 1504.73.
Source 95% interval: 1501.31–1508.15.
Votes: 77636.
Our calculation: the first rating minus the second is 20.49 rating points. Their reported intervals do not overlap. Interval overlap alone is not a pairwise significance test: the covariance and evaluation design matter. This snapshot cannot establish a universal winner or a probability that one model will solve your task.
Decision: keep both candidates if they meet your constraints. Use preference evidence for shortlisting conversational systems, then compare errors on your own prompts. A greater vote count does not guarantee that the voting population or prompts match your users.
Why coding metrics can disagree
Below are the first two complete coding records in source order from our saved LiveBench snapshot, release 2026-06-25, retrieved 2026-10-06T09:00:12.143Z. This is an arithmetic demonstration, not an editorial top-two selection. Full model identifiers retain the source's effort and thinking configuration.
claude-opus-4-5-20251101-thinking-64k-high-effort
Code completion: 80.44.
Code generation: 78.87.
claude-opus-4-6-thinking-auto-high-effort
Code completion: 76.09.
Code generation: 80.28.
Our calculation: first record minus second record = 4.35 points for code completion and -1.41 points for code generation. A positive difference favors the first record on that metric; a negative difference favors the second. These differences use the unrounded source values. They are score-point differences, not percentages of productivity improvement.
Decision: use the metric closest to your workflow. Filling an existing function and producing a solution from a specification are different tasks. Neither raw metric measures your repository's end-to-end bug-fixing success, and neither is the site's aggregated Coding category score. Do not drop the effort suffix and assume the same result applies to a cheaper or faster configuration.
What a verified benchmark name actually means
The first record in the saved SWE-bench Verified submission snapshot is ornith-ai/Ornith-1.5-397B, reporting 86.00 % resolved. Community submission — not marked verified. Its evidence is the linked Ornith-1.5-397B model card. Snapshot retrieved: 2026-10-06T09:00:11.687Z.
Interpretation: “Verified” in the benchmark name identifies the benchmark subset; it does not certify every submitted score. The separate submission flag above is what our source supplies. Even a source-verified submission is not a result independently reproduced by Model VS.
Decision: open the submitted evidence and check the exact model, agent scaffold, tool access, test split, attempt count, and inference budget before using it in a shortlist. If those details are absent, record them as unknown. Do not treat a system's resolved-issue percentage as the probability of fixing your next production bug.