Chart system · archive through 2025-08-29

Three ways to read the same AI race.

Each view preserves providers as the protagonists, uses one green highlight, and keeps every claim inside the verified archive window.

Chatbot Arena archive

Sampled provider frontier ratings, May 2023 to August 2025

Meta highlighted. 827 to 1288 published rating; sampled peak 1396.

Sampled provider frontier ratings over time Google leads the final sampled snapshot at 1466.2 on August 29, 2025. Meta moved from 827.0 to 1288.0 across the selected snapshots and peaked at 1395.9. 800 1000 1200 1400 2023-05 2023-10 2024-03 2024-08 2025-02 2025-08 Google 1466 Mistral 1433 xAI 1431 OpenAI 1429 DeepSeek 1424 Anthropic 1417 Meta 1288
Eleven selected published snapshots are shown. LMArena changed its rating method during this window, including the move from Elo to Bradley-Terry in January 2024 and later style-control and frequency-reweighting changes. The selected snapshots cross LMArena rating-method changes, including Bradley-Terry, style control, and frequency reweighting. Endpoint differences are not normalized like-for-like performance changes. This legacy series is never joined to the current table. Each raw snapshot has a published SHA-256 receipt.

Variation 03 · gap chase

How far did each lab sit from first?

Meta began 447 rating points behind the May 2023 sampled leader and finished 178 points back in August 2025. Because LMArena changed methods during the window, these endpoint gaps are descriptive, not a like-for-like gain.

Leader0
Google 0
Mistral -33
xAI -36
OpenAI -37
DeepSeek -42
Anthropic -50
Meta -178

Variation 04 · reign tape

Who held the archive lead?

Each segment marks the leading provider at one of the 11 selected snapshots. This is snapshot tenure, not a claim about every day between them.

2023-05 OpenAI
2023-07 Anthropic
2023-10 Anthropic
2023-12 OpenAI
2024-03 OpenAI
2024-06 OpenAI
2024-08 OpenAI
2024-11 OpenAI
2025-02 Google
2025-05 Google
2025-08 Google

Snapshot leads: OpenAI 6 · Google 3 · Anthropic 2

Variation 05 · head to head

Anthropic vs OpenAI, snapshot by snapshot.

Anthropic +192

Endpoint difference across sampled ratings; methods changed.

VS
OpenAI +155

Endpoint difference across sampled ratings; methods changed.

Derived values: Anthropic +192; OpenAI +155 rating points. These are not normalized across methodology changes.