Compare Start here

AnthropicLanguage modelSuperseded

Claude Sonnet 4.5

Claimed the 'best coding model in the world' title in autumn 2025.

Compare head-to-head Race it in 3D

Overall Intelligence

#36

of 56
15.1 pts behind #1

0

races led

120

weeks at #1

73.7

score

Released: Sep 29, 2025; 1.1 yr ago
Context: 1M; 64K max output
Price: $3 / $15; per 1M in/out · $6.00 blended
Speed: 80 tok/s; 2.2s to first token
Modalities: text · image; outputs text
Weights: Closed; API and apps

The 30-second version

The autumn-2025 Sonnet that first showed mid-tier models could lead coding benchmarks.

Key takeaways

1Led coding races in late 2025.
2Strong computer use for its time.

Where it ranks

Current snapshot · illustrative scores · click any race to watch it

Sonnet 4.5Best on each axis

Overall

36Overall+15.1 pts

Coding

30Coding+14.7 pts 28Frontend & UI+79 Elo 28Backend & APIs+12.3 pts 32Distributed Systems+16.7 pts 30Full-Stack Apps+14.4 pts 31Debugging & Repair+15.8 pts 26Agentic Coding+15.1 pts 34Code Review & Security+14.3 pts 35Data Science & SQL+14.3 pts 26Mobile Apps+10.8 pts 28Terminal & DevOps+17.3 pts 37Competitive Programming+487

Hard Reasoning

38Hard Reasoning+18.3 pts 41Frontier Questions+33.9 pts 37Abstract & Novel Reasoning+27.9 pts 40Logic & Puzzles+16.8 pts 31Planning & Strategy+14.2 pts

Math

41Math+20.2 pts 43Competition Math+16.3 pts 42Research Math+27.0 pts 38Applied & Quantitative+15.4 pts

Science

37Science+16.7 pts 34Graduate Science QA+10.0 pts 36Biology & Medicine+14.9 pts 40Physics & Chemistry+15.3 pts 35Research Assistance+14.6 pts

Writing & Language

34Writing & Language+15.9 pts 35Creative Writing+96 Elo 34Technical Writing+12.8 pts 34Editing & Style+11.5 pts 38Multilingual+11.5 pts 38Instruction Following+12.9 pts 34Conversation & EQ+94 Elo

Multimodal

35Multimodal+29.5 pts 35Image Understanding+17.6 pts 34Documents & Charts+16.3 pts 35Video Understanding+29.1 pts 39Speech & Audio+40.7 pts

Tool Use & Agents

29Tool Use & Agents+14.6 pts 34Tool Use+12.1 pts 27Computer Use+19.3 pts 36Web Research+19.2 pts 30Long-Horizon+23.5 hrs

Long Context

36Long Context+19.0 pts 371M Recall+15.9 pts 38Long Reasoning+16.7 pts

Speed & Cost

30Speed & Cost+42.0 pts 27Output Speed+440 tok/s 18Latency+1.76s 47Cost-Efficiency+56.8 pts

Trust & Factuality

36Trust & Factuality+17.4 pts 37Factuality+13.9 pts 36Honesty & Calibration+14.7 pts

Overall Intelligence over time

#1 frontierSonnet 4.5V4.1 FlashGLM-5.3

Update history

The distilled version first; open any release for the deeper breakdown.

The Claude Sonnet family

3.7 SonnetFebruary 2025 Sonnet 4.5September 2025 Sonnet 5.5September 2026

60-second briefing

Sonnet 4.5, explained

AI video summary · placeholder render from the auto-written script

From Anthropic's own guide, distilled

Prompt engineering and Claude best practices

Be explicit about the outcome you want; newer Claude models follow instructions literally rather than guessing at hidden intent.
Explain why a rule exists. Context about the goal lets the model generalise the rule to cases you did not list.
Wrap documents, examples and data in XML-style tags so instructions and material never blur together.
Put long documents at the top of the prompt and your question at the end; recall and grounding both improve.
For agentic tasks, state when to act versus when to ask, and ask for a short plan before large edits.

All official guides, distilled →

About Anthropic

Safety-focused lab building the Claude family of models.

Optimising for

Reliable long-horizon agentic work (especially coding and computer use), honest and calibrated answers, and strong writing, with safety research shipped alongside capability.

San Francisco, US · founded 2021

Closest rivals

Sonnet 4.5 is 36th in overall intelligence; leader is Claude Opus 5.5.