Sonnet 5.5 high vs Opus 5.5 low: which uses less of your plan?

Both model names are official and both are listed on Anthropic's pricing page.

low: “Most efficient. Significant token savings with some capability reduction.”

high: “Spends as many tokens as the task needs for excellent results.”

“Effort is a behavioral signal, not a strict token budget. At lower effort levels, Claude still thinks on sufficiently difficult problems, but thinks less than it would at higher effort levels for the same problem.”

On Claude Opus 5.5, adaptive thinking is always on, “so effort is the primary control for how much the model reasons and what a request costs.”

“The tokens Claude spends reasoning are billed as output tokens, even when the thinking text isn't returned to you.”

Anthropic publishes no fixed number or ratio of tokens between low and high effort. not published

On Claude plans, “your usage is affected by several factors, including the length and complexity of your conversations, the features you use, which Claude model you're chatting with, and the effort level you've selected”, and “Higher effort uses more tokens.” No per-model or per-effort rate is published for plans.

OpenAI has no Sonnet/Opus-style names. Its models page calls GPT-6 Astra “Our most capable model for the most demanding work” and GPT-6.1 Sol “Near-Astra performance for complex work at a lower cost”, so we pair those two.

Standard tier, prompts up to 272K input tokens; longer prompts are priced higher.

reasoning.effort “guides the model on how much to think”; “Lower effort favors speed and lower token usage”

Models reason adaptively, “using fewer tokens for simpler tasks and thinking harder for complex tasks.”

Reasoning tokens “still occupy space in the model's context window and are billed as output tokens.”

OpenAI publishes no fixed ratio of tokens between low and high reasoning effort. not published

Google's current Pro model in the API is a preview (gemini-3.1-pro-preview). Google calls Gemini 4 Argon, not a Pro model, “Our flagship reasoning model”, so we show it under the pair.

Paid tier, standard. Gemini 3.1 Pro prices are for prompts up to 200K tokens; longer prompts are priced higher.

Google sets thinking with thinking_level; its thinking page gives no numeric thinking budget for these models.

Guidance: “Use minimal or low thinking for fact retrieval or classification”, and to cut cost lower thinking_level (low or medium) rather than setting a small max_output_tokens.

“When thinking is turned on, response pricing is the sum of output tokens and thinking tokens.” You pay for the full thought tokens, not the returned summary.

Google publishes no token count or multiplier per thinking level. not published

xAI names no mid model. Its models page calls grok-4.7 “the most capable model we've built”; grok-4.5 and grok-4.6 cost the same as grok-4.7. The closest lower-priced current model is grok-4.3, which xAI gives no description.

Prompts under 200K tokens; a prompt that reaches 200K is billed at the higher rate for all its tokens.

xAI's reasoning guide lists reasoning_effort for grok-4.5, grok-4.6 and grok-4.7, not for grok-4.3. not published

low: “Uses some reasoning tokens, but still fast” high: “Uses more reasoning tokens for deeper thinking”

“Reasoning cannot be disabled.”

“When you use a reasoning model, the reasoning tokens are billed as part of your total consumption.” The guide doesn't say at which rate.

xAI publishes no fixed ratio of reasoning tokens between effort levels. not published