Which reasoning effort and model use the fewest tokens?
Use the default effort for everyday work (Medium on Codex, high on Claude Code) and raise it only for hard problems. When tokens are tight, drop one level and one model size. Higher effort is slower, uses more tokens and does not guarantee a better result.
Codex
Start at the lowest effort that gets the result. Stay on the default, Medium, for everyday work, and turn Fast mode off when tokens are tight.
Type /model in the CLI to change the model and the effort together: Low, Medium (default), High, Extra high, Max, Ultra. OpenAI says to use the lowest effort that produces the result you need; higher levels take longer and use more tokens, and Ultra also runs subagents, which costs more than a single agent. Fast mode only buys speed: about 1.5x faster, at 2.5x the usage on the current main models. OpenAI also suggests a smaller model for routine tasks, which stretches the same allowance further.
What the official docs say
Use the lowest reasoning effort that produces the result you need. Increase it for tasks that need more planning, analysis, or checking.Higher reasoning effort can improve results for complex tasks, but it takes longer and uses more tokens.Higher effort can use more of your allowance and does not always produce a better result.Different models can use different amounts of your allowance for the same task. Larger inputs and outputs, higher reasoning settings, Fast mode and tasks with multiple steps can also increase usage.Switch to a smaller model for routine tasks. Using GPT-5.6 Terra or GPT-5.6 Luna can extend your local-message usage limits, depending on the model you switch from.For GPT-5.6, GPT-5.5, and GPT-5.4, Fast mode increases model speed by 1.5x. GPT-5.6 and GPT-5.5 consume credits at 2.5x the Standard rate; GPT-5.4 consumes credits at 2x the Standard rate.Speed configurations increase credit consumption for all applicable models, so they also use included limits faster.Claude
The default, high, is enough for everyday work. To save tokens, set /effort medium, use Sonnet day to day, and keep Opus or Fable for hard problems.
Claude Code has five effort levels: low, medium, high, xhigh and max. The default is high. The docs describe medium as trading some intelligence for fewer tokens, and max as prone to overthinking with diminishing returns. Thinking tokens are billed as output tokens, including the collapsed part you never see. Anthropic suggests Sonnet for most coding and Opus for complex architecture or multi-step reasoning; Fable models cannot turn thinking off and use the weekly limit faster than other models.
What the official docs say
The effort level controls how much thinking Claude applies to a response. Higher effort means more thorough responses, but they take longer and use more tokens, so you'll reach your usage limits faster.Reduces token usage for cost-sensitive work that can trade off some intelligenceCan improve performance on demanding tasks but may show diminishing returns and is prone to overthinking.Sonnet handles most coding tasks well and costs less than Opus. Reserve Opus for complex architectural decisions or multi-step reasoning.Thinking tokens are billed as output tokens, and the default budget can be tens of thousands of tokens per request depending on the model.You are charged for all thinking tokens generated, even when collapsed or redacted.You can use up to 50% of your weekly usage limits on Fable models at no extra cost. They draw from your plan's regular weekly usage limits and use them faster than other Claude models.For Claude Code users on subscription plans (Pro/Max/Team/Enterprise), fast mode is available via usage credits only and not included in the subscription rate limits.Switching model mid-session resets the cache
In Claude Code, each model has its own cache. After you switch with /model, the next message reads the whole conversation again with no cache hits, so a switch deep into a long session is expensive. On most models, changing the effort level does the same, and Claude Code asks you to confirm first. The newest Fable model on a subscription is the exception: changing effort keeps the cache.
Pick the model and the effort at the start of a session, or right after /clear. opusplan switches model every time you enter or leave plan mode, so each switch starts the cache over.
OpenAI does not document whether switching models mid-session affects the Codex cache.
What the official docs say
Each model has its own cache. Switching with /model means the next request reads the entire conversation history with no cache hits, even though the content is identical.On most models, changing the effort level mid-session means the next request reads the entire conversation history with no cache hits. While the cache is still warm, Claude Code asks you to confirm the change first.Common misconceptions
- The highest effort gives the best result
Not necessarily. OpenAI says higher effort can use more of your allowance and does not always produce a better result. Anthropic describes max as showing diminishing returns and being prone to overthinking. Use the default for everyday work and raise it only for hard problems.
- Collapsed thinking is not counted
It is. Claude bills thinking tokens as output tokens, including collapsed or redacted ones. OpenAI lists higher reasoning settings among the things that increase Codex usage.
- Fast mode is only faster, not more expensive
No. Codex Fast mode is about 1.5x faster, and the current main models use your allowance at 2.5x the standard rate. On a Claude subscription, /fast is not covered by your plan limits at all: it is billed to usage credits.
Tips
- Codex: type /model and keep the effort on Medium. Go to High or above only for tasks that need multi-step planning and repeated checking.
- Codex: to pin the default, add model_reasoning_effort = "medium" to ~/.codex/config.toml.
- Codex: type /fast status to see whether Fast mode is on. If tokens are tight, type /fast off.
- Claude Code: type /effort and choose medium. Enter saves it as your default; s applies it to this session only. Typing /effort medium works like Enter.
- Claude Code: when a single prompt needs deeper reasoning, write ultrathink in that prompt instead of raising the effort for the whole session.
- Claude Code: use /model sonnet day to day, and switch models at the start of a session or after /clear.
- Both: give simple subagents a smaller model of their own.