/ LLM benchmarks / live
Live LLM benchmark: new models and ranking changes
What moved on Code Arena, Text Arena and LiveBench. Every model that entered a leaderboard, every new #1 and every overtake inside a top 10, read from the leaderboards twice a day.
Last crawl per source
The crawl is scheduled for 06:00 and 18:00 UTC and often finishes hours later. The table shows when each run fetched. The source date is the leaderboard's own update, which can be older than the crawl.
| Source | Fetched | Source date | Models |
|---|---|---|---|
| Text Arena | 2 Oct 2026 | 413 | |
| Code Arena | 1 Oct 2026 | 138 | |
| LiveBench | release 2026-06-25 | 58 |
Twice a day, scheduled 06:00 and 18:00 UTC
New #1s
Every change at the top of a leaderboard in the last 90 days.
- Gemini 4 Argon High (Google) took #1 on Text Arena
- Claude Opus 5.5 High (Anthropic) took #1 on Text Arena
- Claude Opus 5.5 Max (Anthropic) took #1 on Code Arena
- Claude Fable 5 High (Anthropic) took #1 on Text Arena
- GPT-6 Astra Max (OpenAI) took #1 on Code Arena
- Claude Fable 5.1 Max (Anthropic) took #1 on Code Arena
- Claude Opus 5 Max (Anthropic) took #1 on Code Arena
- Kimi K3 (Moonshot AI) took #1 on Code Arena
- GPT-5.6 Sol xHigh (codex-harness) (OpenAI) took #1 on Code Arena (from #2)
Ranking changes, day by day
The last 30 days. A climb counts when a model overtakes another one inside a top 10 and its rank changes. Models that fell behind are counted, not listed.
- GPT-6.1 Sol Max (OpenAI): Text Arena entered at #21
- Claude Sonnet 5.5 xHigh (Anthropic): Text Arena entered at #45
- Step 5 Preview (Stepfun): Text Arena entered at #74
- Gemini 3.8 Flash High (Google): Text Arena climbed from #11 to #8
- 3 models moved down
- Claude Sonnet 5.5 xHigh (Anthropic): Code Arena entered at #3
- Gemini 4 Argon High (Google): took #1 on Text Arena, Code Arena entered at #8
- GPT-6.1 Sol Max Effort (OpenAI): LiveBench Data Analysis entered at #2, LiveBench Language entered at #2, LiveBench Reasoning entered at #2, LiveBench Math entered at #4, LiveBench Overall entered at #5, LiveBench Instruction Following entered at #12, LiveBench Coding entered at #17, LiveBench Agentic Coding entered at #24
- GPT-6.1 Sol Max (OpenAI): Code Arena entered at #3
- Claude Sonnet 5.5 High (Anthropic): Code Arena entered at #4
- Muse Spark 1.3 Max (Meta): Text Arena climbed from #9 to #8
- 3 models moved down
- Claude Opus 5.5 High (Anthropic): took #1 on Text Arena
- MiMo V2.6 Pro (Xiaomi): Text Arena entered at #23
- DeepSeek V4.1 Flash Max (DeepSeek): Text Arena entered at #29
- GPT-6 Sol Max (OpenAI): Text Arena entered at #60
- MiMo V2.6 Flash (Xiaomi): Text Arena entered at #71
- GPT-6 Luna Max (OpenAI): Text Arena entered at #86
- Grok 4.7 xHigh (SpaceXAI): Text Arena entered at #92
- 2 models moved down
- Step 5 Preview High (Stepfun): Code Arena entered at #29
- Claude Opus 5.5 Max (Anthropic): took #1 on Code Arena
- MiMo V2.6 Pro (Xiaomi): Code Arena entered at #19
- GPT-6 Luna Max (OpenAI): Code Arena entered at #24
- 1 model moved down
- GPT-6 Sol Max (OpenAI): Code Arena entered at #4
- 2 models moved down
- Grok 4.7 xHigh (SpaceXAI): Code Arena entered at #10
- Qwen 3.8 Max (Alibaba): Code Arena climbed from #6 to #4
- Claude Opus 5 High (Anthropic): Code Arena climbed from #7 to #6
- 3 models moved down
- Claude Fable 5 High (Anthropic): took #1 on Text Arena, Code Arena entered at #10
- Muse Spark 1.2 (xHigh) (Meta): Text Arena climbed from #5 to #4
- Claude Opus 4.7 (Anthropic): Text Arena climbed from #8 to #7
- Gemini 3.8 Flash High (Google): Text Arena climbed from #11 to #9
- 3 models moved down
- Muse Spark 1.3 Max (Meta): Text Arena entered at #7
- DeepSeek V4.1 Flash Max (DeepSeek): Code Arena entered at #16
- GPT-6 Astra Max (OpenAI): Text Arena entered at #24
- Claude Opus 4.7 High (Anthropic): Text Arena climbed from #4 to #3
- Muse Spark 1.1 (Meta): Text Arena climbed from #10 to #8
- 2 models moved down
- Muse Spark 1.3 Max (Meta): Code Arena entered at #8
- 2 models moved down
- GPT-6 Astra Max (OpenAI): took #1 on Code Arena
- Muse Spark 1.3 (xHigh) (Meta): Code Arena entered at #11
- Claude Fable 5 (Anthropic): Code Arena climbed from #8 to #7
- Qwen 3.8 Flash Next (Alibaba): Code Arena climbed from #10 to #9
- 4 models moved down
Oldest recorded change: 9 Apr 2026.
Need a model picked for your use case?
Benchmarks rank models on other people's tasks. For a business workflow I test the shortlist on your own data, costs and privacy requirements, and recommend one.