Oleg K

[Oleg K @CheckAllNews]

@scaling01 I would say closer to sonnet 4.6, rather than opus 4.5. Doing much more thinking than the 4.5 models

[Oleg K @CheckAllNews]

@OpenAI Why Think is from the capital letter? 👀

[Oleg K @CheckAllNews]

@ajambrosino min or pro. So if it’s only for pro plan, then probably pro.

[Oleg K @CheckAllNews]

@d4m1n Both skills and instruction about docs should coexist

When both used results even better with real world task. With skill you have more granular control over context.

Opus way better than sonnet with skills, so next model will improve even more.

[Oleg K @CheckAllNews]

@s_streichsbier Codex price is approximately the same as Opus in reality.

Codex is doing a lot of research, which costs more.

Using both Codex for bug researches and doing deep researches and Opus for everything else.

[Oleg K @CheckAllNews]

@mitsuhiko Anthropic should make implicit cache and do not charge extra for a cache.

OpenAI has 24h implicit cache for free.

[Oleg K @CheckAllNews]

@aisdk @zirkelc_ when upgrade to v6 for ai-retry? 😶‍🌫️

[Oleg K @CheckAllNews]

@aisdk @pontusab when upgrade ai-sdk-tools to V6? 👀

[Oleg K @CheckAllNews]

@jidefr @lgrammel Anthropic handles both url and inline data. And it works in ai sdk

OpenAI handles both, but it’s only available as inline data in ai sdk

Google vertex can handle both, but only handles inline in ai sdk(uses google ai provider)

Gemini API can handle only inline data per doc

[Oleg K @CheckAllNews]

@is_mohitsingh @joshtriedcoding Per my observation it increases token usage by 30%.

[Oleg K @CheckAllNews]

@aidenybai Gemini 3 is doing the same, and I hate it.

[Oleg K @CheckAllNews]

@haider1 130 IQ 🤣

[Oleg K @CheckAllNews]

@claudeai When Opus 4.5?

[Oleg K @CheckAllNews]

@OfficialLoganK Overall, it's a great model, but it still has the same problem as with Gemini 2.5.

Sometimes it's not calling tools and printing results as plain text.

[Oleg K @CheckAllNews]

@wartranslated so why do not send their troops to fight in Ukraine?

[Oleg K @CheckAllNews]

@steipete This one more interesting 😂

[Oleg K @CheckAllNews]

@nicoalbanese10 And sometimes gpt is going insane 😂

[Oleg K @CheckAllNews]

@nicoalbanese10 GPT-5 high for complex/big tasks or problems that CC can not solve. Especially resolving types in TS.

GPT-5 too slow for small quick edits.

CC for anything else.

Using like this for a week.

CLI/Vs code codex extension is unusable when you want to accept edits manually.

[Oleg K @CheckAllNews]

@karthikkalyan90 @aisdk @DSPyOSS @shadcn Can it optimize descriptions for a tool and input schema descriptions?

[Oleg K @CheckAllNews]

@sama When in API?