DEEPSEEK · MODEL RELEASE TRACKER

DeepSeek-V3.1

DeepSeek-V3.1 is a released DeepSeek model version (announced Aug 21, 2025) that introduces an agent-focused workflow and hybrid inference modes. It emphasizes improved agent/tool use, multi-step reasoning, updated tokenizer/chat templates, and open-source weights; a subsequent update (Terminus) further improved stability and language consistency.

UPDATED 2026-09-26

CURRENT SNAPSHOT2/5 DIMENSIONS WITH DATA

The dimensions that change the decision.

PRICING

Not established from the available sources.

CONTEXT WINDOW

Not established from the available sources.

MODALITIES

Not established from the available sources.

BENCHMARKS

'Think' mode speed vs. earlier modelDeepSeek-V3.1-Think is reported to reach answers in less time versus DeepSeek-R1-0528.2025-08-21

AVAILABILITY

API format support (Anthropic)Supports the Anthropic API format.2025-08-21

RELEASE TIMELINE

Published, source-backed release events only.

2025-08-21

1 SOURCES · IMPORTANCE 7.0

DeepSeek releases DeepSeek‑V3.1 with hybrid 'Think & Non‑Think' agent mode and open-source weights

DeepSeek released DeepSeek‑V3.1, introducing a hybrid inference mode called Think & Non‑Think (toggleable via the “DeepThink” button at chat.deepseek.com), faster 'Think' responses vs. DeepSeek‑R1‑0528, improved tool use and multi-step agent skills, Anthropic API format support, and beta Strict Function Calling support. V3.1 Base received 840B tokens of continued pretraining for long-context extension; the tokenizer and chat template were updated (tokenizer_config.json); open-source weights for DeepSeek‑V3.1 and DeepSeek‑V3.1‑Base are published on Hugging Face, and new pricing/off-peak discount changes take effect Sep 5, 2025, 16:00 UTC.

→

VERIFIABLE FACTS

Every value stays attached to a source and date.

RELEASE · Initial release / announcementDEVELOPER CLAIM

Announced on 2025-08-21 on DeepSeek's website as 'DeepSeek-V3.1'.

RELEASE · Update / 'Terminus' releaseDEVELOPER CLAIM

Updated/renamed to 'DeepSeek-V3.1-Terminus' (published 2025-09-22); the Terminus update was posted with open-source weights on Hugging Face.

CAPABILITIES · Hybrid inference modes ('Think' and 'Non-Think')DEVELOPER CLAIM

Introduces hybrid inference with two modes — 'Think' and 'Non-Think' (user can toggle via the DeepThink button).

BENCHMARKS · 'Think' mode speed vs. earlier modelDEVELOPER CLAIM

DeepSeek-V3.1-Think is reported to reach answers in less time versus DeepSeek-R1-0528.

CAPABILITIES · Agent skills, tool use, and multi-step reasoningDEVELOPER CLAIM

Post-training and updates are described as boosting tool use, stronger multi-step agent task performance, and stronger multi-step reasoning for complex search tasks; Code Agent and Search Agent performance explicitly called out as improved.

AVAILABILITY · API format support (Anthropic)DEVELOPER CLAIM

Supports the Anthropic API format.

WHAT CHANGED

Stored passport versions, without reconstructed history.

2026-09-26Passport created6 facts

DeepSeek releases DeepSeek‑V3.1 with hybrid 'Think & Non‑Think' agent mode and open-source weights

Announced on 2025-08-21 on DeepSeek's website as 'DeepSeek-V3.1'.

Updated/renamed to 'DeepSeek-V3.1-Terminus' (published 2025-09-22); the Terminus update was posted with open-source weights on Hugging Face.

Introduces hybrid inference with two modes — 'Think' and 'Non-Think' (user can toggle via the DeepThink button).

DeepSeek-V3.1-Think is reported to reach answers in less time versus DeepSeek-R1-0528.

Post-training and updates are described as boosting tool use, stronger multi-step agent task performance, and stronger multi-step reasoning for complex search tasks; Code Agent and Search Agent performance explicitly called out as improved.

Supports the Anthropic API format.

Optional analytics