The week in AI models.
Sep 21–27, 2026. 18 new releases, 7 new leaders on tests and 3 models joining the capability index.
Highlights
Counted from this week’s releases, new leaders and price changes.
Top new releaseClaude Opus 5.5Anthropic, 167.3 on the capability index
Most releasesOpenAI4 new models- Most new top spotsClaude Opus 5.5New leader on 6 tests
New releases
Models from known makers that the maker sells, several hosts list or a test covers, highest on the capability index first, then the most widely offered.
| Model | Released | Capability index | Price | Context |
|---|---|---|---|---|
| Claude Opus 5.5Anthropic | Sep 22, 2026 | 167.3 | $8.00 | 1M |
| GPT-6 SolOpenAI | Sep 22, 2026 | 162.7 | $4.00 | 1.1M |
| GPT-6 LunaOpenAI | Sep 22, 2026 | 156.3 | $0.20 | 1.1M |
 Grok 4.7xAI | Sep 21, 2026 | 153.5 | $3.00 | 500K |
 MiMo-V2.6-FlashXiaomi | Sep 22, 2026 | Not on the index yet | $0.175 | 1M |
| MiMo-V2.6-ProXiaomi | Sep 22, 2026 | Not on the index yet | $0.544 | 1M |
| MiMo-V2.6-Pro-UltraSpeedXiaomi | Sep 21, 2026 | Not on the index yet | $5.44 | 1M |
 Qwen 3.8 Max PrimeAlibaba | Sep 23, 2026 | Not on the index yet | $6.00 | 1M |
 Solar Mini 4Upstage | Sep 23, 2026 | Not on the index yet | $0.0875 | 524.3K |
| GPT-6 Luna ProOpenAI | Sep 22, 2026 | Not on the index yet | $0.20 | 1.1M |
| GPT-6 Sol ProOpenAI | Sep 22, 2026 | Not on the index yet | $4.00 | 1.1M |
 MiniMax-M3.1-Flash-PreviewMiniMax | Sep 27, 2026 | Not on the index yet | No paid listing | 1M |
 LongCat 2.5 PreviewMeituan | Sep 25, 2026 | Not on the index yet | $0.525 | 1M |
| LongCat 2.5 Preview FreeMeituan | Sep 25, 2026 | Not on the index yet | No paid listing | 1M |
| MiMo V2.6 Flash UncensoredXiaomi | Sep 25, 2026 | Not on the index yet | $0.75 | 1M |
 GLM 5.3 PrimeZ.ai | Sep 23, 2026 | Not on the index yet | $4.30 | 1M |
| Claude Opus 5.5 FastAnthropic | Sep 22, 2026 | Not on the index yet | $16.00 | 1M |
 Command A+Cohere | Sep 22, 2026 | Not on the index yet | $0.60 | 192K |
New leaders
- New leaderWithin margin
Claude Opus 4.6 took the top spot on Arena coding at 1551, from Claude Fable 5.
- New leaderWithin margin
Claude Opus 5.5 took the top spot on Arena text at 1509, from Claude Fable 5.
- New leaderWithin margin
Claude Opus 5.5 took the top spot on Arena hard prompts at 1541, from Claude Opus 4.6.
- New leaderWithin margin
Claude Opus 5.5 took the top spot on Arena creative writing at 1521, from Claude Fable 5.
- New leaderWithin margin
Claude Opus 5.5 took the top spot on Arena instruction following at 1516, from Claude Opus 4.6.
- New leader
Claude Opus 5.5 took the top spot on LiveBench coding at 89.3%, from Claude Fable 5.1.
- New leader
Claude Opus 5.5 took the top spot on LiveBench mathematics at 97.1%, from Claude Fable 5.1.
Price changes
No maker changed a listed price this week.
Joined the capability index
Joined the indexamazon--nova-pro joined the capability index at 123.8.
Joined the indexLlama 3 70B Instruct joined the capability index at 122.9.
- Joined the index
Llama 3 8B Instruct joined the capability index at 116.3.
Releases, prices and context windows from models.dev (MIT); LiveBench (Apache-2.0); Epoch AI (CC BY 4.0); Arena (CC BY 4.0). A week runs Monday to Sunday, UTC. Changes count from when the site began keeping this record, as on What changed. RSS feed