Theo
Theo is a content creator focused on technology topics, primarily producing content and commentary in the areas of artificial intelligence, programming tools, and software engineering. His output covers AI-assisted programming, developer tool evaluations, and analysis of relevant technical research. He has established a notable presence in the technology community, particularly through in-depth discussions of cutting-edge AI programming practices.
Core Facts
AI inference prices, measured per unit of compute, have dropped approximately 10-fold over the past two years, a phenomenon the industry calls the AI inference price spiral.
The reviewer recommends letting Opus 5.5 handle implementation and long-horizon building, while assigning GPT-4.1 Sol to review, root cause analysis, codebase investigation, and task triage.
Sonnet 5.5 is on par with Opus 5.5 in offensive and defensive cybersecurity capabilities, making it the first model in the Sonnet series to adopt equivalent cybersecurity protection mechanisms.
Timeline (last 90 days)
After testing, programming blogger Theo pointed out that more than half of Gemini 4's judgments about his codebase were wrong, making it the only one of the 7 tested models with more than half errors.
Developer Theo placed WonderFull in his S-tier models, tied with Opus 5.5.
Theo said on a livestream himself that buying Fable-level intelligence at the price of GLM 5.3 Flash 'will absolutely happen'.
Two years ago Theo made a video titled 'AI will not keep improving', and now admits he was wildly wrong, having thought at the time that development had stalled because old benchmarks could not measure the capabilities of new models.
The sharp increase in tokens generated per prompt is a manifestation of people using models for larger and harder tasks after they became cheaper, better, and more capable, and rising bills and productivity gains are related.
In developer Theo's model ranking, Opus 5.5 is S-tier, Muse Wonder is A-tier, and GPT-6.1 Sol is also in the A-tier.
Measuring AI cost should consider three dimensions: how many tokens can be purchased per dollar, how many tokens are consumed to complete a task, and how much it costs to solve a problem at a specific level of intelligence.
The author considers GLM 5.3 Flash to be a better model than GPT-5 in its own right, with stronger visual processing and better tool calling, particularly crushing GPT-5 in performance during hours-long agentic workflows.
Theo speculates that the fundamental divide between OpenAI and Anthropic is a difference in engineering versus research culture: OpenAI is engineer-led, while Anthropic places comparatively less emphasis on software engineers
During the Sonnet 4 era, Anthropic heavily used models to generate code, accumulating a codebase full of slop, and is now using stronger models to go back and clean it up
40 more timeline events
All Facts (20)
AI inference prices, measured per unit of compute, have dropped approximately 10-fold over the past two years, a phenomenon the industry calls the AI inference price spiral.
65%UnverifiedTheo认为如果信息在代码库中能找到就不应放在AgentMD里，因为模型已经非常擅长通过bash命令和工具探索代码库
90%UnverifiedTheo建议在AgentMD中写入指令让Agent在遇到令其惊讶的事情时提醒开发者，目的是利用Agent反馈发现代码库中的架构问题
90%Unverified在AgentMD中告诉模型不要使用某技术（如tRPC），模型反而会更多地想到该技术，类似于'粉红大象'效应
70%Partially VerifiedThe reviewer recommends letting Opus 5.5 handle implementation and long-horizon building, while assigning GPT-4.1 Sol to review, root cause analysis, codebase investigation, and task triage.
65%Partially VerifiedSonnet 5.5 is on par with Opus 5.5 in offensive and defensive cybersecurity capabilities, making it the first model in the Sonnet series to adopt equivalent cybersecurity protection mechanisms.
65%UnverifiedAfter testing, programming blogger Theo pointed out that more than half of Gemini 4's judgments about his codebase were wrong, making it the only one of the 7 tested models with more than half errors.
50%UnverifiedDeveloper Theo placed WonderFull in his S-tier models, tied with Opus 5.5.
50%UnverifiedThe sharp increase in tokens generated per prompt is a manifestation of people using models for larger and harder tasks after they became cheaper, better, and more capable, and rising bills and productivity gains are related.
50%UnverifiedIn developer Theo's model ranking, Opus 5.5 is S-tier, Muse Wonder is A-tier, and GPT-6.1 Sol is also in the A-tier.
50%UnverifiedThe author considers GLM 5.3 Flash to be a better model than GPT-5 in its own right, with stronger visual processing and better tool calling, particularly crushing GPT-5 in performance during hours-long agentic workflows.
50%UnverifiedMeasuring AI cost should consider three dimensions: how many tokens can be purchased per dollar, how many tokens are consumed to complete a task, and how much it costs to solve a problem at a specific level of intelligence.
50%UnverifiedMetrics-driven optimization carries risks: agents targeting reductions in React commits or V8 call counts tend to cut network requests that don't affect interactivity, leading to overly aggressive caching with insufficient invalidation
50%UnverifiedDuring the Sonnet 4 era, Anthropic heavily used models to generate code, accumulating a codebase full of slop, and is now using stronger models to go back and clean it up
50%UnverifiedTheo speculates that the fundamental divide between OpenAI and Anthropic is a difference in engineering versus research culture: OpenAI is engineer-led, while Anthropic places comparatively less emphasis on software engineers
50%UnverifiedTheo found through real-world testing that Claude.ai has a data consistency issue: after deleting a session in one browser, refreshing another browser still displays the deleted session, and clicking on it throws an error without re-validating the sidebar
50%UnverifiedTheo recommends avoiding the low and Max tiers, suggesting the medium tier is generally sufficient.
50%UnverifiedSonnet 5.5 is notably inferior to Opus on frontend design and far inferior to Fable, while Theo considers Fable 5.1 to be the best design model.
50%UnverifiedIn Theo's testing, Sonnet 5.5 averaged approximately 150 TPS via the official subscription, while Opus averaged approximately 100 TPS.
50%UnverifiedOn the codebase audit benchmark task, Sonnet's cost was approximately half that of Opus but scored slightly higher, achieving the best cost-per-point, and Sonnet took only about 5 minutes while Opus took nearly twice as long.
50%