Get the most from CoreCode
Choosing a Claude model
Claude Code can use the Opus, Sonnet, Haiku and Fable families. The right one can halve the cost of the same work; the wrong one is either expensive or keeps getting it wrong. The model descriptions and price relations here come from Anthropic's model configuration guide and official pricing.
This page covers Claude only. The GPT, Gemini and Grok models CoreCode also serves are in Supported models.
- Sonnet 5.5 for everyday coding, Opus 5.5 for complex reasoning and architecture, Haiku 5.5 for simple tasks, and Fable 5.1 only for the hardest, longest tasks.
- At official prices, Opus 5.5 costs 2 times Sonnet 5.5 and Fable 5.1 5 times; Haiku 5.5 costs a twentieth of Sonnet, rising to a quarter once a prompt passes 100,000 tokens.
opusplanuses Opus in plan mode and switches to Sonnet to execute.- The 5.5 models default to
mediumeffort; change it with/effort. Higher is slower and costs more. - Pick the model when a task starts: a mid-session switch rewrites the whole context into the cache at the write price on the next turn.
What each family is for
| Alias | Currently | For |
|---|---|---|
sonnet | Sonnet 5.5 | Everyday coding: fixing bugs, writing tests, building a clearly specified feature |
opus | Opus 5.5 | Complex reasoning: architecture, refactors across many files, problems Sonnet keeps getting wrong |
haiku | Haiku 5.5 | Simple, fast tasks |
fable | Fable 5.1 | The hardest and longest-running tasks |
opusplan | Opus + Sonnet | Opus designs the plan in plan mode, then Sonnet writes the code |
Aliases move to new versions over time. To pin one, use the full model name, such as claude-opus-5-5, or set ANTHROPIC_DEFAULT_OPUS_MODEL, ANTHROPIC_DEFAULT_SONNET_MODEL and the like.
Each model's full specs, prices and changes from the previous generation are in Claude Fable 5.1, Claude Opus 5.5 and Claude Sonnet 5.5.
How the prices compare
Relative official prices, with Sonnet 5.5 as 1. CoreCode's prices are in the pricing table, in the same proportions.
| Model | Input and output price | Cache read (against its own input price) |
|---|---|---|
| Haiku 5.5 | 1/20; 1/4 for prompts over 100,000 tokens | 0.1× |
| Sonnet 5.5 | 1 | 0.05× |
| Opus 5.5 | 2× | 0.05× |
| Fable 5.1 | 5× | 0.025× |
Mind Haiku's long-context price: after a while, a Claude Code session easily sends more than 100,000 tokens of context per turn, and then Haiku costs five times its base price. Haiku suits short tasks, subagents and one-off calls from scripts better than long sessions.
How to choose
- Start with Sonnet. It handles most coding well at half the price of Opus.
- Switch to Opus when stuck. When Sonnet misses two or three times, or the work is a cross-module design, use Opus. Getting it right once on Opus beats reworking on Sonnet.
- Get both with
opusplan. Let Opus settle the plan in plan mode (Shift+Tab), then execution runs on Sonnet. - Keep Fable for genuinely hard work. At five times Sonnet, a long task adds up.
Claude Code's default model depends on the account type; Anthropic's guide gives Opus 5.5 for the Anthropic API. If you are unsure what you are on, run /status or /model; to default to Sonnet, see switching and pinning.
Effort level
Effort sets how deeply the model thinks at each step. Low is fast and cheap for straightforward work; high thinks harder for hard problems, but is slower and uses more tokens. Opus 5.5, Sonnet 5.5 and Haiku 5.5 support low, medium, high, xhigh and max, and default to medium.
- Run
/effortin a session; confirming saves it as that model's default. maxapplies to the current session only, unless set through theCLAUDE_CODE_EFFORT_LEVELenvironment variable.- You can also start with
claude --effort high.
Switching and pinning models
| How | Effect |
|---|---|
/model in a session | Pick a model; later sessions start on it |
claude --model sonnet | This launch only |
"model": "sonnet" in settings.json | Every session defaults to Sonnet |
ANTHROPIC_DEFAULT_OPUS_MODEL and similar | Pin an alias to a specific version |
New models need a recent Claude Code: Opus 5.5 from v2.1.280, Sonnet 5.5 from v2.1.284, Haiku 5.5 from v2.1.293. If you get "does not support this model", run claude update.
What a mid-session switch does to the cache
The prompt cache is per model. Switch halfway through a session and the new model cannot read the earlier cache: the next turn rewrites the whole context at the write price and is noticeably slower (see Anthropic's Claude Code caching guide). So:
- Pick the model at the start of a task and finish the task on it.
- If you must switch, follow it with a substantial piece of work to spread the rewrite, or
/compactfirst so there is less to rewrite.
More on caching in Caching and billing and Spending less on Claude Code.
Through a relay, is it the model you chose?
Anthropic's guide says it plainly: ANTHROPIC_BASE_URL changes where requests are sent, not which model answers them. Choose Opus on CoreCode and the request goes to Opus. To check for yourself, look at each request's model and price in the console's usage logs, or verify the upstream is Anthropic with the checking script.
FAQ
How much more does Opus cost than Sonnet?
At official prices, Opus 5.5's input and output cost 2 times Sonnet 5.5's, with the same cache-read multiplier. CoreCode's prices are in the pricing table.
Haiku is cheapest, so why not use it for everything?
Capability: Haiku gets complex tasks wrong more often, and rework costs more. And long context: past 100,000 tokens of prompt, Haiku costs five times as much, a length long Claude Code sessions easily reach.
What is opusplan?
A model setting that uses Opus for reasoning and design in plan mode, then switches to Sonnet to write the code. Turn it on with /model opusplan.
Is a higher effort level always better?
No. Higher effort thinks harder but is slower and uses more tokens. The default medium, or even low, is enough for straightforward work; raise it for hard problems.