OpenAI Launches GPT-5.1-Codex-Max with Exceptional Value for Money

AIbase基地

Published inAI News · 3 min read · Dec 5, 2025

Recently, OpenAI officially launched its latest programming model, GPT-5.1-Codex-Max, for developers. This new model not only shows significant performance improvements but also maintains the same price as its predecessor, GPT-5, becoming a hot topic among developers.

The main highlights of GPT-5.1-Codex-Max include its "agent-style" coding capabilities and the ability to run for extended periods. Through a technology called "compaction," this model can work continuously within a context of more than one million tokens, making it more stable when performing complex tasks. According to internal tests, GPT-5.1-Codex-Max can run continuously for over 24 hours, marking a major advancement in the field of programming.

In several authoritative benchmark tests, GPT-5.1-Codex-Max performed exceptionally well. For example, in tests such as SWE-Bench Verified, SWE-Lancer IC SWE, and TerminalBench2.0, the model achieved scores of 77.9%, 79.9%, and 58.1%, respectively, showing a clear improvement compared to the standard version of GPT-5.1-Codex. These test results show that GPT-5.1-Codex-Max has a greater advantage when handling complex, multi-step development tasks.

In terms of commercial use, OpenAI has integrated GPT-5.1-Codex-Max into its API at the same price as GPT-5. The input cost is $1.25 per million tokens, and the output cost is $10 per million tokens. This move undoubtedly represents OpenAI's further expansion into the developer market, especially targeting programmers who use Windows as their primary development platform. Unlike previous Codex series models that were optimized for Unix environments, this new model was specifically adapted for Windows during the training phase.

Currently, GPT-5.1-Codex-Max has quickly entered the ecosystems of several popular development tools, including Cursor, GitHub Copilot, and Linear, among other programming assistants. In GitHub Copilot, this model will be available to Pro, Pro+, Business, and Enterprise subscription users, providing programmers with a more efficient coding experience.

OpenAI's Fourth Acquisition This Year! $400 Million in Stock Acquires Neptune for Training Monitoring Stack Enhancement

OpenAI acquired Polish startup Neptune in an all-stock deal valued below $400 million. This marks OpenAI's fourth acquisition of 2025, aimed at integrating Neptune's model monitoring and debugging capabilities into the GPT training infrastructure. Neptune services will be gradually phased out before Q1 2026, with the entire team joining OpenAI.

DeepSeek Penguin Family Leaked at Midnight: 512 Budget Emperor Targets GPT-5.2, No Reasoning Mumble Challenges Rapid Response

The new model matrix codenamed 'Penguin' by OpenAI has been exposed, including the flagship Emperor512, mid-range Rockhopper64, lightweight Macaroni16, and zero-reasoning Mumble0 with four levels of reasoning budget, covering all scenarios from cloud to edge. The flagship Emperor512 has a reasoning budget of 512 units, 8-10 times that of current models, and may become the core of GPT-5.2.

OpenAI, DeepMind Lag Behind in AI Safety! Authoritative Institution Releases AI Safety Index, Calls for Urgent Industry Corrections

"The AI Safety Index 2025" report shows that leading AI companies have collectively failed in survival risk control and safety process disclosure. The industry's average score is only 42/100, far below the EU standard. Only three companies have disclosed systematic risk identification methods, while OpenAI, DeepMind, and others have not revealed measures against "superintelligence".

OpenAI Launches 'Confession' Framework: Making AI More Honest and Brave Enough to Admit Mistakes!

OpenAI has launched a 'Confession' training framework aimed at improving the honesty of AI models. This mechanism requires models to actively acknowledge their mistakes or inappropriate behaviors after providing the main answer, correcting issues that may arise in traditional training such as concealing the truth or giving inaccurate responses.

AI Investment Bubble Alert: Anthropic CEO Warns of Excessive Market Risks

The CEO of Anthropic warned that AI industry investments are overheated, pointing out that some companies have committed to investing billions of dollars to develop AI systems, but face significant financial risks. The industry faces a dilemma: building advanced AI requires substantial funding, but excessive investment may lead to a bubble.