Claude Code Model Selection: Opus 5.5, Sonnet 5, Fable 5.1, and Haiku

The best Claude model for coding is Claude Opus 5.5, and since Claude Code v2.1.280 it is also the model you get without choosing one. Opus 5.5 shipped on September 22, 2026 at $4 per million input tokens and $20 per million output, 20% below Opus 5, and it leads both Opus 5 and Fable 5.1 on all nine benchmarks in Anthropic's launch table. Anthropic's models overview now opens with one instruction: "If you're unsure which model to use, start with Claude Opus 5.5 for most workloads."

Claude Code model selection now comes down to three decisions: which model answers, how hard you tell it to think, and how fast the choice spends your weekly allowance. Opus 5.5 moved two of those at once. It defaults to medium effort instead of high, and its cache reads cost $0.20 per million tokens, exactly what Sonnet 5 charges. This guide covers all three, plus the configuration details that quietly send work to the wrong model.

/model claude-opus-5-5

The Current Claude Code Lineup

Four models are worth knowing by name. Everything else in the picker is a previous generation you can still pin, including Opus 5, which Anthropic now lists as legacy at $5/$25.

Sonnet 5's $2/$10 launched as introductory pricing through August 31, 2026. Anthropic's pricing page now lists it as the standard rate and states that the scheduled move to $3/$15 "will not occur." So on the rate card Opus 5.5 costs exactly 2x Sonnet 5 on input and output and 1x on cache reads, and Fable 5.1 costs 2.5x Opus 5.5. Those three ratios decide most routing questions on this page.

Two things about the Haiku row are easy to miss. The haiku alias is documented only as "the fast and efficient Haiku model," with no version guarantee, so pin claude-haiku-4-5 when the version matters. And Haiku 4.5 is a different shape of model from the other three: a 200K context window, not 1M, 64K max output, a February 2025 reliable knowledge cutoff, no effort-level support, and the nearest retirement floor on the current lineup: not sooner than October 15, 2026.

What Model Does Claude Code Use by Default?

Opus 5.5, on every subscription plan and on the Anthropic API. The default model setting is not itself a model alias. It clears any override and reverts to the recommended model for your account type:

This is the largest routing change in the release. Before v2.1.280, default resolved to Sonnet 5 on Pro and Team Standard and to Opus 5 on Max, Team Premium, Enterprise, and the API. Pro and Team Standard users now start every session on Opus without touching a setting.

Three things override that table. An Enterprise admin can set an organization default model, which default resolves to instead. A model value you saved in your settings wins over the account default, so a settings file that still pins claude-opus-5 keeps you on the legacy model at $5/$25. And Fable models are not the account-type default on any plan or provider. Run /status to see which model a session is actually on.

Model Aliases

Claude Code provides model aliases so you do not need exact version numbers. These are the shortcuts you use daily:

best is the alias almost nobody uses and probably should. It resolves to Fable 5.1 where your seat has Fable access and to the latest Opus otherwise, which makes it the one setting that stays correct across a mixed team. The fable alias itself moved to Fable 5.1 in v2.1.257, except in Claude apps gateway sessions, where fable and best still resolve to Fable 5.

The Aliases Do Not Mean the Same Thing on Every Provider

This is the detail that quietly breaks routing advice copied between teams. opus and sonnet point at the newest model your provider offers, not the newest model Anthropic has shipped:

So --model sonnet on Bedrock gets you Sonnet 4.5, two generations behind the same flag on the Anthropic API, and --model opus on Foundry gets you Opus 4.6. Where an alias resolves to an older model, select the full model name or set ANTHROPIC_DEFAULT_OPUS_MODEL or ANTHROPIC_DEFAULT_SONNET_MODEL. Aliases update over time by design, so pin the full name whenever you need a specific version:

/model claude-opus-5-5
/model claude-sonnet-5
/model claude-fable-5-1

Version floors matter more than usual, because an out-of-date client hides models rather than erroring: Opus 5.5 needs Claude Code v2.1.280 or later, Fable 5.1 needs v2.1.257, Opus 5 needs v2.1.219, and Sonnet 5 needs v2.1.197. Run claude update if a model you expect is missing from the picker.

The 1M-token context window is standard now. On the Anthropic API, Fable 5.1, Fable 5, Sonnet 5, and Opus 4.7 and later run with the 1M window on every plan, including Pro, with no [1m] variant to select and no usage credits to enable. The window carries no pricing premium beyond 200K. Opus 4.6 and Sonnet 4.6 still reach 1M only through their [1m] variants.

Effort Defaults Changed With Opus 5.5

Every model that supports effort defaults to high in Claude Code, with two exceptions: Opus 5.5 defaults to medium, and Opus 4.7 defaults to xhigh. The medium default is not a cost cut dressed up as a feature. On Terminal-Bench 4.0, Opus 5.5 at medium beats Opus 5 at max for about a fifth of the cost per attempt, and on FrontierCode its medium score of 54.6% beats every setting on every model in the launch table.

Two details catch people upgrading. A top-level effortLevel saved in your user settings before Claude Code stored effort per model does not apply to Opus 5.5, so an old xhigh preference becomes medium after the switch until you pick a level with /effort or the /model picker. And on Opus 5.5 with an API key or a Claude subscription, changing effort mid-session keeps the prompt cache, where on most models it forces a full uncached re-read. That makes per-task effort changes cheap on Opus 5.5 in a way they are not on Sonnet 5. The Opus 5.5 best practices guide covers which level to use per task, and the effort ladder guide covers what each rung costs.

How to Choose a Claude Model for Coding

Most developers still route by an old price ratio. Opus once cost five times Sonnet, so rationing it made sense. The rate card now says 2x, and even that overstates the gap on agentic work, because the line that dominates an agent's bill is priced the same on both models.

What the 2x Rate Card Costs on a Real Agent Session

Take the coding agent from our Opus 5.5 pricing breakdown: a 200K-token cached prefix carried through 100 tool-calling turns, adding about 2K new input tokens and 1.5K output tokens per turn. That is 20M cached tokens re-read, 200K new input tokens written to the 5-minute cache, and 150K output tokens.

At identical token counts, Opus 5.5 costs about 1.33x Sonnet 5 on this session, not 2x, because cache reads are half of the Opus bill and cost the same on both. New input is billed once, at the cache-write rate. This is our illustration, not Anthropic's, and it holds token counts equal on purpose. In practice Opus 5.5 runs at medium by default and Sonnet 5 at high, and partner reports on Opus 5.5 describe fewer turns per task, so the real gap on hard work is likely smaller still. A chat workload that never caches pays the full 2x.

Opus 5.5: The Default for Coding

Best for: most real development work

claude --model claude-opus-5-5

Use Opus 5.5 for:

• Agentic coding sessions that run unattended for more than a few minutes
• Large-scale refactoring across multiple systems
• Complex debugging with intricate dependencies
• Architectural decisions requiring deep reasoning
• Security reviews and code auditing, with the classifier caveat covered under fallback below

Why it is the default: On Anthropic's launch table, Opus 5.5 scores 66.4 on Terminal-Bench 4.0 against 55.8 for Fable 5.1 and 52.3 for Opus 5, and 57.8 on CursorBench 4.0 against 51.8 and 46.6. GitHub's Mario Rodriguez reported that in VS Code "it solved more terminal tasks than Opus 5 in less than half the steps." Anthropic also reports output more than 30% faster than Opus 5.

Cost reality: $4/$20 is 2x Sonnet 5 on the rate card and about 1.3x on a cache-heavy agent session. On a subscription it is the model your plan gives you by default.

One prompt change matters for routing. Opus 5.5 thinks on every request and cannot have thinking disabled, so effort is the only control. If your prompts still carry "think step by step" or "think hard" instructions, delete them. They stack on top of behavior the model already has and cost tokens for no quality gain.

Sonnet 5: The Volume Play

Best for: work where output price per call is the deciding variable

claude --model claude-sonnet-5

Use Sonnet 5 for:

• Mechanical passes: applying a spec verbatim, string swaps, config churn
• Sub-agents doing execution while a stronger model plans
• High-frequency automation and batch jobs, where Sonnet 5's Batch API rate is $1/$5
• Test writing, documentation, and routine feature work

Where it sits: Sonnet 5 scores 63.2% on SWE-bench Pro against Opus 4.8's 69.2%. Anthropic has published no Sonnet 5 versus Opus 5.5 head-to-head. Opus 5 beat Opus 4.8, and Opus 5.5 beats Opus 5 on all nine launch rows, so assume the gap to Opus 5.5 is wider than the six points Sonnet 5 gives up to Opus 4.8. One more gap matters for coding: Sonnet 5's reliable knowledge cutoff is January 2026, against June 2026 for Opus 5.5, which shows up on libraries released in between.

Sonnet 5 is the right model when that capability gap costs you less than the output bill does. That usually means sub-agents, and the sub-agent best practices guide covers how to set their model.

Fable 5.1: The Measured Exception

Best for: workloads you have benchmarked and found Opus 5.5 short on

/model fable

Fable 5.1 costs $10/$50, 2.5x Opus 5.5, and it trails Opus 5.5 on every row of the Opus 5.5 launch table. Anthropic keeps it for "demanding reasoning and long-horizon agentic work, or when your evals on Claude Opus 5.5 at higher effort still fall short," and concedes the gap is "narrower than these scores suggest." It is the only generally available Mythos-class model, and the long-horizon scientific research results from its September launch are the territory where testing both still makes sense.

Two operational details decide it more often than any score. Fable 5.1 carries mandatory 30-day data retention unless Anthropic expressly authorizes otherwise, while Opus 5.5 is available under zero data retention. And on subscriptions Fable 5.1 draws on a capped share of your allowance or on usage credits, covered in the next section. Treat it as a workload-specific exception you have measured, not a general upgrade.

Haiku 4.5: The Speed Tier

When to downgrade: simple, repetitive tasks

claude --model haiku

Use Haiku 4.5 for:

• Simple file reads and formatting
• Basic syntax validation
• Quick status checks
• Text transformations and background functionality

Warning: Haiku struggles with complex logic and multi-file projects, and its 200K window is a fifth of the other three. Its real job on a modern setup is background work and the ANTHROPIC_DEFAULT_HAIKU_MODEL slot, not your main thread.

Opusplan: Plan Expensive, Execute Cheap

When to use: sessions that mix heavy planning with mechanical implementation

claude --model opusplan

How it works:

• In plan mode: uses opus, which is Opus 5.5 on the Anthropic API, for reasoning and architecture decisions
• In execution mode: switches to sonnet for code generation and implementation

This is the model-tier split expressed as a single alias: the expensive model decides, the cheap model types. It suits refactoring sessions, feature planning, and any workflow built on planning mode. On the Anthropic API both phases run with the 1M window. Elsewhere, /model opusplan[1m] requests it for both phases (v2.1.265 or later).

One cost is easy to miss. Each model has its own prompt cache, so every plan-mode toggle under opusplan is a model switch that re-reads the whole conversation uncached. Toggle once per task, not every few turns. The usage optimization guide has the numbers on when the split pays.

Your Weekly Allowance Is Part of the Decision

On a subscription, model selection has a quota dimension on top of quality and price.

Opus 5.5 is now the default on Pro and Team Standard. Our read: at twice Sonnet 5's per-token rate, it will use plan limits faster, so on those plans the upgrade trades headroom for capability. If you hit limits sooner than you did last week, /model sonnet puts you back where you were. Anthropic raised five-hour usage limits on Pro, Max, Team, and seat-based Enterprise plans alongside the launch, which offsets part of the difference.

Fable 5.1 carries a ceiling the others do not. On Max plans and premium Team and Enterprise seats, it is included up to 50% of your weekly usage limits. On Pro plans and standard seats it runs on usage credits at API rates, and Claude Code shows a consent prompt before a Fable request bills them.

Three consequences follow:

1. Fable 5.1 can run out while Opus 5.5 still has headroom. Hitting the 50% cap ends your Fable access, not your week. /model claude-opus-5-5 is the recovery.
2. Fable costs 2.5x on whichever meter you are billed by. On the API that is the token rate. On a Max plan it shows up as a heavier draw per session against the same weekly bucket.
3. Opus 5.5 has no per-model ceiling. It draws on your weekly allowance like any other model.

The Fable usage-credits guide covers the credit mechanics and the terminal messages at the cap. For plan-level limits, see higher usage limits.

Tactical Model Switching

The ClaudeFast approach: start on Opus 5.5 at medium, drop down for volume, and switch at task boundaries rather than mid-task.

During Development Sessions

# Start your session on the current default
claude --model claude-opus-5-5
 
# Raise effort for one hard task. On Opus 5.5 this keeps the cache
/effort xhigh
 
# Drop to Sonnet for a long mechanical pass
/model claude-sonnet-5
 
# Back to Opus for the hard part
/model claude-opus-5-5
 
# Drop to Haiku for trivial file operations
/model haiku

Every /model switch costs one slow, uncached turn, because the new model reads the full conversation with no cache hits. While the cache is still warm, Claude Code asks you to confirm the switch. Changing effort on Opus 5.5 does not carry that cost, so reach for /effort first when the question is depth rather than price.

Cost Optimization Pattern

1. Default: Opus 5.5 at medium, because on a subscription it is the plan default and on a cache-heavy API session it costs about 1.3x Sonnet 5
2. Tune: raise effort per task before changing models
3. Hybrid: use opusplan for sessions that mix planning and implementation
4. Delegate: put Sonnet 5 on sub-agents doing execution while the main thread plans
5. Downgrade: Haiku for file operations and background work
6. Escalate: Fable 5.1 only on workloads you have measured, with the 50% weekly cap in view
7. Monitor: run /status and /usage to see which tier is consuming your allowance and whether switches are costing cache misses

The savings come from not running Opus on work a cheaper tier finishes identically, which is a routing problem, not a rationing one.

Model Selection by Task Type

Reasoning Quality Rankings

1. Opus 5.5: ahead of every other Claude model on all nine rows of Anthropic's September 22 launch table
2. Fable 5.1: behind Opus 5.5 on published scores, kept for long-horizon research work Anthropic says the scores understate. See how to prompt Fable once you have picked it
3. Sonnet 5: strong for most development scenarios, below Opus 4.8 on SWE-bench Pro
4. Haiku 4.5: basic reasoning only

Cost Efficiency Rankings

1. Haiku 4.5: cheapest per task, limited capability and a 200K window
2. Sonnet 5: best throughput per dollar on high-volume, low-cache work
3. Opus 5.5: best result per dollar on anything hard, and near Sonnet 5 on cache-heavy agent loops
4. Fable 5.1: 2.5x the Opus 5.5 rate for a narrow research edge

Speed Rankings

1. Haiku 4.5: fastest in Anthropic's lineup table, and shallow
2. Sonnet 5: fast, with good depth
3. Opus 5.5: "moderate" latency, with output more than 30% faster than Opus 5. Optiver reported it matched Opus 5's quality "in about half the turns, time and output tokens," which can make it faster end to end
4. Fable 5.1: slowest in the lineup

How to Change the Model in Claude Code

Claude Code gives you several ways to set your model, with a clear priority chain. Higher-priority settings override lower ones:

1. During session: /model <alias|name> switches models mid-session (highest priority)
2. At startup: claude --model <alias|name> sets the model for that session
3. Environment variable: ANTHROPIC_MODEL=<alias|name> persists across sessions
4. Settings file: the model field in settings.json for permanent configuration (lowest priority)

To set a default model, run /model claude-opus-5-5 inside a session, which saves it to your user settings, or add the model field to ~/.claude/settings.json yourself:

{
  "permissions": {},
  "model": "claude-opus-5-5"
}

Leaving model unset follows the account default, which already resolves to Opus 5.5 on every subscription. Pin it only when you need a specific version to hold across future releases. ClaudeFast's Code Kit includes model routing configuration that applies these selection rules automatically, so mechanical work lands on cheaper tiers and heavy work escalates without a decision on every prompt.

Controlling Which Models Aliases Map To

To override which model an alias points to, for example to pin a version or use a Bedrock or Google Cloud deployment, use these environment variables:

The deprecated ANTHROPIC_SMALL_FAST_MODEL variable has been replaced by ANTHROPIC_DEFAULT_HAIKU_MODEL.

These are especially useful for Bedrock, Foundry, and Google Cloud users who need inference profile ARNs, deployment names, or version names instead of standard Anthropic model IDs.

CLAUDE_CODE_SUBAGENT_MODEL changed meaning in v2.1.251. It used to override everything, including a sub-agent definition's model frontmatter. It is now a default: Claude Code resolves a sub-agent's model from the per-invocation model parameter first, then the definition's model field (where inherit means the main conversation's model), then this variable, then the main conversation's model. It also does not change the built-in Explore and Plan sub-agents on its own. To put every sub-agent on one tier regardless of their definitions, set both variables (v2.1.257 or later):

{
  "env": {
    "CLAUDE_CODE_SUBAGENT_MODEL": "sonnet",
    "CLAUDE_CODE_SUBAGENT_MODEL_FORCE": "1"
  }
}

Run /tasks while a sub-agent is running to see which model it actually landed on. The sub-agent best practices guide covers how to pick the value.

When Claude Code Switches Models on You

Two mechanisms move you off your chosen model, and neither is triggered by usage limits.

Content-based fallback. Fable models, Opus 5.5, and Opus 5 run safety classifiers that most often flag cybersecurity and biology content. On Fable 5.1, Fable 5, and Opus 5.5, biology-flagged requests re-run on Opus 5 and cybersecurity-flagged requests re-run on Opus 4.8. On Opus 5, cybersecurity-flagged requests re-run on Opus 4.8, while biology-flagged requests end in a refusal. After a fallback the session stays on the fallback model, so run /model to return. Anthropic launched Opus 5.5 with what it calls "a temporarily wider safety margin against jailbreaks," so security teams moving from Opus 5 should expect more cyber fallbacks for now.

Availability-based fallback chains. Configure these with --fallback-model sonnet,haiku or a fallbackModel array in settings. They fire when the primary model is overloaded, unavailable, or returns another non-retryable server error, and the switch lasts for the current turn only. Authentication, billing, rate-limit, request-size, and transport errors never trigger a switch.

The practical consequence for routing: a repository containing security or biology material can trip a classifier on the first request of a session, because that request carries your CLAUDE.md and git status. If a session keeps landing on a fallback model, claude --safe-mode disables customizations so you can tell whether your own config is the trigger.

Checking Your Current Model

You can see which model you are using in two places:

• Run /status to display your current model and account information
• Configure a StatusLine to show the model persistently in your terminal

Prompt Caching Configuration

Claude Code uses prompt caching automatically. If you need to disable it for debugging or benchmarking, set one of these environment variables to 1:

Disabling prompt caching raises cost and latency, so only do it with a specific reason.

Caching matters more for model selection than it used to. Opus 5.5 bills cache hits at 0.05x base input, half the 0.1x multiplier every earlier Opus used, which is how its $0.20 cache read matches Sonnet 5. Anthropic's note on why that line matters: cache reads "make up the majority of agentic and coding work costs." The minimum cacheable prompt stays at 512 tokens.

Common Model Selection Mistakes

Mistake 1: Routing by the rate card alone

• Problem: the 2x input and output ratio makes Opus 5.5 look twice the price of Sonnet 5, but cache reads cost the same on both, and they dominate agent sessions
• Solution: default to Opus 5.5 and drop to Sonnet 5 where output volume, not capability, is the constraint

Mistake 2: Treating Fable 5.1 as the upgrade path

• Problem: 2.5x the Opus 5.5 rate, a capped share of your weekly allowance, and behind Opus 5.5 on every published launch benchmark
• Solution: benchmark your own workload first. If Opus 5.5 at higher effort matches it, the answer is Opus 5.5

Mistake 3: Pinning a legacy model in settings

• Problem: a settings file that pins claude-opus-5 overrides the new default, so you pay $5/$25 for a model Opus 5.5 beats on every launch benchmark
• Solution: pin claude-opus-5-5, or remove the model key and follow the account default

Mistake 4: Assuming CLAUDE_CODE_SUBAGENT_MODEL overrides your agent definitions

• Problem: since v2.1.251 a sub-agent's own model field wins, so agents pinned to opus or inherit keep billing at the main-thread rate
• Solution: set CLAUDE_CODE_SUBAGENT_MODEL_FORCE=1 alongside it, or change the model field in the definitions themselves

Mistake 5: Switching models mid-task

• Problem: every /model switch re-reads the whole conversation uncached, and opusplan does it on every plan-mode toggle
• Solution: switch at task boundaries, and on Opus 5.5 change effort instead when you only need more depth

Mistake 6: Staying on Haiku too long

• Problem: poor code quality, missed issues, and a 200K window on large repositories
• Solution: move to Sonnet 5 or Opus 5.5 for any real development

Frequently Asked Questions

What is the best Claude model for coding?

Claude Opus 5.5. It leads Fable 5.1 and Opus 5 on all nine benchmarks in Anthropic's September 22, 2026 launch table, including 66.4 on Terminal-Bench 4.0 against Fable 5.1's 55.8, at $4/$20 per million tokens, less than half Fable 5.1's rate. It is also the default model in Claude Code on every subscription plan.

Which Claude model should I use in Claude Code?

Opus 5.5 for anything that requires judgment, Sonnet 5 for volume and for sub-agents doing execution, Haiku 4.5 for file operations and background work, and Fable 5.1 only for a workload you have benchmarked and found Opus 5.5 short on. For most people the answer is to leave the default alone and drop down deliberately.

What model does Claude Code use by default?

Opus 5.5 on Pro, Max, Team, Enterprise, the Anthropic API, Claude Platform on AWS, Amazon Bedrock, and Google Cloud's Agent Platform, from Claude Code v2.1.280. Microsoft Foundry defaults to Sonnet 4.5. An Enterprise organization default or a model value in your settings overrides it.

Is Opus 5.5 better than Sonnet 5 for coding?

Yes. Anthropic has not published a direct head-to-head, but Sonnet 5 trails Opus 4.8 by six points on SWE-bench Pro, and Opus 5.5 beats Opus 5, which beat Opus 4.8. Sonnet 5 costs half the Opus 5.5 rate on input and output and the same on cache reads, so it wins on high-volume work where output price decides.

How do I change the model in Claude Code?

Run /model claude-opus-5-5 mid-session, or claude --model claude-opus-5-5 at startup. Typing /model with a model name also saves it as your default for new sessions, or you can set the model field in your settings file. Session-level choices always win over the settings file.

Does Claude Code pick a model automatically?

Partly. The default alias resolves to the recommended model for your account type, which is now Opus 5.5 almost everywhere. The best alias resolves to Fable 5.1 where you have Fable access and the latest Opus otherwise. opusplan automates a split rather than a choice: Opus during plan mode, Sonnet during execution. Claude Code also moves you off your model when a safety classifier flags a request, but never because you approached a usage limit.

Is Opus 5.5 worth it over Sonnet 5?

On a subscription it is already your default, so the question is whether the faster draw on your allowance is worth it, and /model sonnet answers it for any session where it is not. On the API it costs 2x Sonnet 5 on input and output and the same on cache reads, about 1.3x on a cache-heavy agent session. Anything where a wrong answer costs more than that difference goes to Opus 5.5.

Which model do sub-agents use?

The per-invocation model parameter if Claude passes one, then the sub-agent definition's model field, then CLAUDE_CODE_SUBAGENT_MODEL, then your main conversation's model. Set CLAUDE_CODE_SUBAGENT_MODEL_FORCE=1 to make the variable win over everything. The sub-agent best practices guide covers how to choose the value.

What happens when I run out of Fable 5.1?

On Max and premium seats, Fable 5.1 is capped at 50% of your weekly usage limits. Hitting that cap ends your Fable access, not your week. Run /model claude-opus-5-5 and keep working. The Fable usage-credits guide covers the messages and the credit path on Pro.

Quick Reference

For beginners: stay on the default. It is Opus 5.5, and it makes fewer mistakes to learn from.

For daily development: Opus 5.5 on the main thread, Sonnet 5 on sub-agents.

For budget optimization: route by phase and by cache profile. Mechanical passes go to Sonnet 5, decisions stay on Opus 5.5, and long cached agent loops are cheaper on Opus than the rate card suggests.

For learning: check our configuration guide to set up model defaults.

The model you pick decides who is answering. The effort level you set decides how long they work on it, and on Opus 5.5 the dial is the cheaper of the two to move.

Next Action: run /status to confirm you are on Opus 5.5, check your settings file for a pinned claude-opus-5, then give your execution sub-agents model: sonnet so delegated work stops billing at the top rate. The usage optimization guide covers what that split saves.

For detailed specs on every Claude model Anthropic has released, see the complete model timeline.

Related Guides:

• Claude Opus 5.5 - benchmarks, pricing, and the four API breaking changes
• Claude Opus 5.5 best practices - effort per task, prompt cleanup, and long runs in Claude Code
• Claude Fable 5.1 - the Mythos-class tier above Opus and when it still earns 2.5x
• Claude Sonnet 5 - the volume tier and its benchmarks against Opus 4.8
• Model versus effort - when to change the model and when to change the dial
• Ultracode and the effort ladder - what each effort level costs
• Deep Thinking Techniques - getting more reasoning out of a session
• Usage Optimization - track and reduce your costs
• Context Management - keep long sessions inside the window you are paying for
• Speed Optimization - what to tune once the model is right
• Efficiency Patterns - build frameworks for consistent output