Claude
Claude Opus 5.5
Public rank 1 · Proprietary
Overall 86
Coding 82.8
Agentic 88.8
Reasoning 86.1
AI Model Comparison
Claude Opus 5.5 vs GPT-6.1 Sol vs Gemini 4 Argon
Latest public provider picks: Claude is Claude Opus 5.5, ChatGPT is GPT-6.1 Sol, and Gemini is Gemini 4 Argon.
A current public benchmark comparison of latest Claude, ChatGPT/GPT, and Gemini models, with what the scores mean for coding, automation, reasoning, and cost per task.
Updated 2026-10-13
Current comparison
Latest public Claude, ChatGPT, and Gemini
The table uses public benchmark rows and excludes restricted-access and unknown-availability models before selecting each provider.
ChatGPT
GPT-6.1 Sol
Public rank 2 · Proprietary
Overall 83.5
Coding 71.1
Agentic 74.4
Reasoning 90.5
Gemini
Gemini 4 Argon
Public rank 6 · Proprietary
Overall 80.9
Coding 71.6
Agentic 74.1
Reasoning 80.6
How to read it
What the scores mean
Overall score is a shortlist signal. Coding, agentic workflow, reasoning, context, and token price decide whether the right model changes by task.
Coding is not one number
A high coding score is the first filter, not the final decision. Repo repair, terminal-agent behavior, long context, and local tests decide whether the model can actually move a codebase.
Agentic work changes the winner
Workflow automation rewards tool selection, persistence, browser or terminal actions, and recovery from bad intermediate results. That can favor a different model than raw coding or reasoning.
Cost means completed task
Token price matters only after accounting for retries, long context, output length, cached input, tool calls, and human correction. Cheap tokens can be expensive if the task fails.
Method
Use this as a model shortlist
For production, choose the model by task: repo repair, automation, retrieval, regulated-domain work, latency, and cost per completed task.
This page is intentionally narrow. It compares the latest public Claude, ChatGPT/GPT, and Gemini rows, then points back to the full benchmark guide for benchmark-family selection, cost interpretation, and local evaluation design.
Read the benchmark guide ->Inspect first
Sources
- BenchLM public leaderboard endpoint
- BenchLM public pricing endpoint
- Models.dev model database
- LLM model benchmarks guide
- LMArena Leaderboard
- SWE-bench repository
- Terminal-Bench 2.1
Third-party data note: live rows come from public benchmark and pricing feeds, not internal Dreamers testing. Restricted-access and unknown-availability rows are excluded from public leaders; public previews remain public. Cost per task estimates use 100K input and 30K output tokens; tools, retries, cache, and failed attempts are excluded.