AI model week: MiniMax M3.1 Flash, DeepSeek V4.1 Flash and 23 more releases
Oct 19, 2026 · 3 min read · weeklyreleasesleaderboardprice-changes
This week ModelBench tracked 25 new model releases, including five provider-confirmed launches. Twelve models changed price, and Gemini 4 Argon moved into the top five. Data as of 2026-10-19.
New releases
25 models were added between October 13 and October 19. Five were confirmed by their providers.
- MiniMax M3.1 Flash — score 67.1, proprietary, input $0.15, output $0.60. Source
- DeepSeek V4.1 Flash 1017 — score 65.67, open, provider-confirmed, input $0.15, output $0.60. Source
- Muse Spark Open — score 62.77, open, input $1.25, output $4.25. Source
- Beam 1.0 501B-A23B — score 51.58, open, provider-confirmed, input $0.30, output $1.00. Source
- Nemotron 3.5 Super VL 120B A12B — score 49.38, open, provider-confirmed. Source
- Nemotron 3.5 Bolt 74B-A9B — score 45.25, open. Source
- LFM2.5-24B-A2B — score 37.71, open, input $0.10, output $1.00. Source
- Mercury 3 — proprietary, input $0.10, output $0.35. Source
- Glimpse — input $0.60, output $3.00. Source
- d1-ultra — proprietary, provider-confirmed, input $0.04. Source
- Pareto 26.10 — input $0.80, output $3.20. Source
- Nemotron 3.5 Super VL — open, input $0.15, output $0.50. Source
- Moondream 3.1 — open, input $0.30, output $2.50. Source
- Moondream 3 Preview — open, input $0.30, output $2.50. Source
- Jais-3-MoVA-36B-A4B — open. Source
- Jais-3-7B — open. Source
- Jais-3-375B-A23B — open. Source
- Jais-3-32B — open. Source
- Jais-3-3.7B — open. Source
- Jais-3-0.9B — open. Source
- Hy-Memory-30B-A3B — open. Source
- Ultramarine 1 Flash — proprietary, provider-confirmed, input $0.15, output $0.50. Source
- Grok Voice 3.0 — proprietary. Source
- Summarize — input $0.04. Source
- Extract — input $0.04. Source
Price changes
Twelve models had price changes this week. Prices are in US dollars per million tokens.
| Model | Input before | Input after | Output before | Output after |
|---|---|---|---|---|
| Hy3 | $0.13 | $0.08 | $0.53 | $0.33 |
| Hy4 preview | $0.83 | $0.75 | $2.50 | $2.25 |
| GLM-5.3-Flash | $0.03 | $0.05 | $0.42 | $0.16 |
| DeepSeek V4 Pro 0813 | $1.32 | $0.66 | $3.96 | $1.98 |
| DeepSeek V4.1 Flash 1017 | $0.30 | $0.15 | $1.20 | $0.60 |
| GLM-5.3 | $0.08 | $0.08 | $4.80 | $1.40 |
| Qwen3 30B A3B Instruct 2507 | $0.10 | $0.09 | $0.30 | $0.30 |
| Gemma 3 27B | $0.08 | $0.08 | $0.45 | $0.35 |
| DeepSeek V4 Pro 0423 | $0.36 | $0.96 | $0.72 | $1.91 |
| Nemotron 3.5 Super VL | $0.10 | $0.15 | $0.40 | $0.50 |
| Beam 1.0 501B-A23B | $1.00 | $0.30 | $3.00 | $1.00 |
| GLM-4.6 | $0.50 | $0.55 | $2.00 | $2.20 |
Leaderboard
Top 5:
- Claude Opus 5.5 — Anthropic — 86.66
- GPT-6.1 Sol — OpenAI — 84.21
- GPT-6 Astra — OpenAI — 84.05
- Claude Fable 5.1 — Anthropic — 83.24
- Gemini 4 Argon — Google — 82.93
Moves inside the top 20:
- Gemini 4 Argon rose from 7 to 5 (score 82.93).
- Claude Sonnet 5.5 fell from 5 to 6 (score 82.22).
- Claude Fable 5 fell from 6 to 7 (score 81.35).
- Claude Opus 4.8 rose from 17 to 16 (score 69.97).
- Kimi K3 rose from 18 to 17 (score 69.95).
- Qwen3.8 Max rose from 19 to 18 (score 69.73).
- Step 5 Preview rose from 20 to 19 (score 69.57).
- GPT-5.5 rose from 21 to 20 (score 69.21).
For the full record, with the source behind every release date, see Model updates.