# Claude Fable 5.2 vs GPT-6.2 Sol: 4x per Token, 12x per Task

> Claude Fable 5.2 costs 4x GPT-6.2 Sol per token and 6.5–12x per task in same-prompt tests, where both passed every answer. Fable leads only from high effort up.

- URL: https://blog.laozhang.ai/en/posts/claude-fable-5-2-vs-gpt-6-2-sol
- Published: 2026-11-21
- Updated: 2026-11-21
- Author: LaoZhang AI Team (https://blog.laozhang.ai/en/about)
- Topic: Model Comparisons
- Tags: Claude Fable 5.2, GPT-6.2 Sol, Claude API, OpenAI API, model comparison, API pricing

---
Move any work that Claude Fable 5.2 (`claude-fable-5-2`) already does well at medium or low effort to GPT-6.2 Sol (`gpt-6.2-sol`) at medium, its default, or high, and keep Fable at high or above for the jobs that need its top scores. Per token, Fable costs 4x Sol: $8 input and $40 output per million tokens against $2 and $10, on both vendors' price lists as of November 18, 2026. Per task the gap is wider: about 12x in three of four same-prompt tests run in November 2026 and 6.5x in the fourth, because Fable also writes three to four times as many output tokens and counts the same text as roughly 1.4–1.5x more input tokens. Every answer in those tests passed on both models, including an eight-task set built to be hard, so what they measured is cost, tokens and time, not quality.

The quality gap comes from benchmarks. On the Artificial Analysis Intelligence Index v5.0.1, as of November 20, 2026, Sol at max scores 57 for $1.01 per task, one point above Fable at medium (56, $3.80), but no Sol setting reaches Fable at high (62, $9.08) or max (65, $26.54). OpenAI's own launch charts agree at the top: Sol ties Fable on Agents' Last Exam, and Fable's top setting leads on the other four charts, though Sol at high beats Fable at high on the CAD one.

Two kinds of limits can decide before quality does. Fable requires 30-day data retention and rejects forced `tool_choice`; Sol bills a whole request at $4 / $15 once input passes 272K tokens and isn't in regular ChatGPT chat. If none of that applies to you, run your own prompts on both and compare cost per passing result.

## Should you move from Fable 5.2 to GPT-6.2 Sol? Yes for medium-effort work

Move a job to Sol when Fable at medium or low effort is good enough for it today, because on Artificial Analysis each of those Fable settings has a Sol setting that scores the same or higher for 17–27% of the cost. Keep a job on Fable when you run it at high, xhigh or max for a reason, such as a hard multi-step agent task or analysis where a few points of accuracy are worth more than the bill, because Sol's highest score sits 5 points below Fable high. OpenAI's changelog frames Sol the same way, as the model "for everyday coding and agent workflows", released November 17, 2026.

| Your work | Start with | Effort | Switch to the other model when |
| --- | --- | --- | --- |
| Code changes, scripts and fixes that tests or exact answers can check | `gpt-6.2-sol` | medium (default) | Sol still fails your checks at xhigh |
| Agent tasks in a coding CLI | `gpt-6.2-sol` in Codex CLI | high | Tasks run long, span many files, and Sol's choices need frequent correction |
| Reading long logs or documents under 272K input tokens | `gpt-6.2-sol` | medium (default) | Inputs are PDFs you don't want to convert, or single requests exceed Sol's 922,000-token input limit |
| Finance, legal, healthcare or strategy analysis judged on accuracy | `claude-fable-5-2` | high | Sol at max passes your review set, since it costs about a ninth of Fable high on Artificial Analysis |
| Hard terminal, science, CAD or automation work at the top score | `claude-fable-5-2` | xhigh or max | Not on current benchmarks; Sol's best trails Fable's best by about 5–10 points |

The first three rows are where the November tests ran, and Sol passed all of them at a fraction of Fable's cost. The last two rows come from benchmarks: on the four business categories Artificial Analysis reports, Fable's best setting beats Sol's best by 8–12 points (finance and accounting 65 vs 57, strategy and operations 67 vs 59, legal 63 vs 55, healthcare 65 vs 53, as of November 20, 2026). The CLI row also carries a warning from people who used both for a full day: the developer posting as kunchenguid on X wrote on November 19 that GPT-6.2 Astra and Sol "are not a drop-in replacement for fable 5.2 and opus 5.5", pointing to their judgment about when to ask the user and when to go ahead. The tests below don't cover that kind of long, interactive session.

If you want OpenAI's model at Fable's own price tier, OpenAI positions GPT-6.2 Astra against Fable, compared in [Claude Fable 5.2 vs GPT-6 Astra vs GPT-6.2 Astra: Cost per Task](https://blog.laozhang.ai/en/posts/claude-fable-5-2-vs-gpt-6-astra). What changed from the previous Sol, and how Sol compares with Anthropic's cheaper flagship, is in [GPT-6.2 Sol vs 6.1 Sol vs Opus 5.5: Benchmarks and Tested Cost](https://blog.laozhang.ai/en/posts/gpt-6-2-sol-vs-gpt-6-1-sol-vs-claude-opus-5-5). If you'd rather cut cost without leaving Claude, see [Claude Fable 5.2 vs Opus 5.5: When Fable High Beats Opus Max](https://blog.laozhang.ai/en/posts/claude-fable-5-2-vs-opus-5-5).

## Fable 5.2 vs GPT-6.2 Sol pricing: 4x per token, about 12x per task

Claude Fable 5.2 lists at exactly 4x GPT-6.2 Sol on standard input and output, 2x on cache hits, and 2–2.67x on requests whose input passes 272K tokens, where Sol moves to a higher rate and Fable stays flat. Rates per million tokens below are from [Anthropic's Claude API pricing page](https://platform.claude.com/docs/en/about-claude/pricing) and [OpenAI's API pricing page](https://developers.openai.com/api/docs/pricing), as of November 18, 2026.

| Rate per 1M tokens | Claude Fable 5.2 | GPT-6.2 Sol | Fable ÷ Sol |
| --- | --- | --- | --- |
| Input / output, standard | $8 / $40 | $2 / $10 | 4x / 4x |
| Cached input (cache hit) | $0.20 | $0.10 | 2x |
| Cache writes | $10 (5-minute), $16 (1-hour) | $2.50 (30-minute), $5.00 (12-hour) | Different windows |
| Input / output, requests over 272K input tokens | $8 / $40 (no step) | $4 / $15 on the whole request | 2x / 2.67x |
| Batch, input / output | $4 / $20 | $1 / $5 | 4x / 4x |
| Faster tiers, input / output | Fast mode $16 / $80 | Fast $4 / $20; Ultrafast $6 / $30 until February 28, 2027, then 6x standard | 4x against Sol Fast |

Per task, the multiple grows for two reasons that stack on top of the token price. Fable writes far more output: on the hard eight-task set described below, it produced 88,074 output tokens at its default effort against Sol's 28,131, 3.13x as many. And Fable counts the same text as more tokens: on one English passage it counted 1.45x the input tokens of GPT-6.2 Astra, and Sol reported exactly the same input totals as Astra on 11 identical prompts, so Sol counted the same as Astra (November 18, 2026). Across the November tests Fable's input counts ran 1.41–1.51x Sol's for identical prompts. Because output tokens dominate the bill, the per-task multiple lands close to 4 × the output ratio: 4 × 3.13 = 12.5, and the actual figure on that set was $3.6118 ÷ $0.2971 = 12.2x.

Artificial Analysis shows the same effect across every effort level. Comparing each setting with Sol's setting of the same name, Fable costs 17.2x at low ($3.10 vs $0.18 per index task), 9.3x at medium ($3.80 vs $0.41), 16.2x at high ($9.08 vs $0.56), 22.8x at xhigh ($14.83 vs $0.65) and 26.3x at max ($26.54 vs $1.01), as of November 20, 2026. The per-token "4 times as much" figure is accurate as a price, but none of these per-task ratios comes close to it. The one place the gap narrows is a long request: above 272K input tokens Sol's whole request costs $4 input and $15 output, so on such a request Fable is 2x on input and 2.67x on output, and with Fable's larger token counts still well above that per task.

## Same prompts on both models, November 2026: every answer passed

Four sets of same-prompt tests, run between November 10 and November 20, 2026, gave the same result: both models passed every graded answer, and at default effort Fable cost 6.5–12.2x as much. The API runs went through one gateway, laozhang.ai, with the same prompts and graders for both models, and every dollar figure below is computed from returned token counts at the vendors' list prices. Fable's rows in the first three tests date from November 10–11, a week before Sol launched, while the hard set ran both models on November 20; Fable's price didn't change in between.

*Disclosure: [laozhang.ai is our own service](https://blog.laozhang.ai/en/about#laozhang). Its prices are platform pricing, not an independent comparison.*

| Test | Claude Fable 5.2 | GPT-6.2 Sol | Fable ÷ Sol |
| --- | --- | --- | --- |
| 8 prompts, 11 graded answers, API defaults | Effort high, November 11: 11/11, $0.7436, 173.2 s | Effort medium, November 18: 11/11, $0.0610, 56 s | 12.2x cost, 3.1x time |
| 146 KB log, seven exact counts, API defaults | Effort high, November 11: all right, $1.02, 73.7 s | Effort medium, November 18: all right, $0.1581, 13.6 s | 6.5x cost, 5.4x time |
| Same 11-task kit in each vendor's coding CLI, effort high | Claude Code, November 10, 2 runs: 22/22, $2.1545 per run, 221.8 s (mean) | Codex CLI 0.166.0, November 20, 3 runs: 33/33, $0.184 per run, 195 s (median) | 11.7x cost, 1.14x time (Fable mean vs Sol median) |
| Hard 8-task set, two samples each, API defaults | Effort high, November 20: 16/16, $3.6118, 25.8 s median per call | Effort medium, November 20: 16/16, $0.2971, 7.3 s median per call | 12.2x cost, 3.27x summed time |

![Four same-prompt tests of Claude Fable 5.2 and GPT-6.2 Sol: both models passed every answer, and Fable cost 6.5x to 12.2x as much per run](https://blog.laozhang.ai/posts/en/claude-fable-5-2-vs-gpt-6-2-sol/img/same-prompt-tests.webp)

The hard set was written to find a quality gap and didn't. Its tasks and hidden tests were frozen at 15:48 UTC on November 20, before any model saw them: a meeting-slot finder with 37 hidden checks, a small integer expression language with operator precedence and short-circuit operators, a scheduler that has to handle daylight-saving gaps and Lord Howe Island's 30-minute shift, a glob matcher with brace expansion and `**`, a bug hunt in an 83-line cache, a JSON plan under eight interlocking rules, and two no-code arithmetic problems with exact answers. Every configuration solved every task: Fable at its default high and at medium ($1.7857), and Sol at medium, high ($0.4574) and xhigh ($0.7008). Raising Sol from medium to xhigh nearly quadrupled its reasoning tokens, from 14,502 to 55,000, and changed no result. Per task, Fable at default cost 9.2x (the JSON plan) to 19.7x (the expression language) as much as Sol at default, and Sol was cheaper on all eight. One hidden glob test checked a rule the prompt never stated and was excluded after the run; every answer from both models had failed only that test, so the comparison didn't change.

A few smaller results matter for the choice of settings. On the 11-answer set, Fable at low effort cost $0.4201, still 6.9x Sol at default, and Sol at high passed 11/11 for $0.0812, so moving Sol up a level costs little. On the hard set, Fable medium against Sol xhigh, the closest pairing on Artificial Analysis, was still 2.55x the cost. In the CLI kit the cost was mostly context rather than answers: each Sol run read about 300,000 input tokens, 88% of them cached and priced at Sol's $0.10 cached rate, which is why Sol medium ($0.176) cost almost the same as Sol high there.

These tests have limits worth stating. They are single-turn API calls and small CLI tasks with one or two samples per cell, not long agent sessions. The CLI row compares two agent tools as well as two models, its dollar figures convert subscription usage at API list price rather than reading a bill, and its Fable column was not re-run after November 10. With every answer passing on both sides, the tests say how much Fable costs to reach the same result on tasks like these; they can't say which model is smarter.

## Which Sol setting matches which Fable setting: none reaches Fable high

Sol max matches Fable medium, Sol xhigh ties it, and Sol high and medium sit around Fable low; nothing on the Sol side reaches Fable high, xhigh or max. The ladder below is the [Artificial Analysis Claude Fable 5.2 vs GPT-6.2 Sol comparison](https://artificialanalysis.ai/models/comparisons/claude-fable-5-2-vs-gpt-6-2-sol) sorted by Intelligence Index v5.0.1 score, with the cost of one index task, the minutes per task and the time to the first answer token, as of November 20, 2026.

| Setting | Index | Cost per task | Minutes per task | First answer token |
| --- | --- | --- | --- | --- |
| Fable 5.2 max | 65 | $26.54 | 42.0 | 262.5 s |
| Fable 5.2 xhigh | 62 | $14.83 | 24.5 | 59.4 s |
| Fable 5.2 high | 62 | $9.08 | 15.2 | 25.1 s |
| GPT-6.2 Sol max | 57 | $1.01 | 9.6 | 83.9 s |
| Fable 5.2 medium | 56 | $3.80 | 6.5 | 11.2 s |
| GPT-6.2 Sol xhigh | 56 | $0.65 | 5.9 | 34.9 s |
| GPT-6.2 Sol high | 55 | $0.56 | 5.0 | 13.5 s |
| Fable 5.2 low | 54 | $3.10 | 5.5 | 8.3 s |
| GPT-6.2 Sol medium | 53 | $0.41 | 3.5 | 3.6 s |
| GPT-6.2 Sol low | 41 | $0.18 | 1.6 | 2.0 s |

Read across rows with similar scores and the effort rule falls out. Sol max scores one point above Fable medium at 27% of the cost ($1.01 ÷ $3.80); Sol xhigh ties Fable medium at 17%; Sol high is one point above Fable low at 18%; and Sol medium is one point below Fable low at 13%. So if you run Fable at low today, Sol at high is the like-for-like replacement, and if you run Fable at medium, Sol at xhigh or max is. Above that there's no Sol replacement at all: Sol's ceiling of 57 is 5 points below Fable high and 8 below Fable max. The index ranks settings on its nine evaluations; it doesn't predict any given workload, which is why the November tests above, all passed by both, don't contradict it. Unrounded, the same scores appear on OpenAI's own chart of this index: Fable 53.9 to 64.6 and Sol 40.6 to 57.3 from low to max.

![Artificial Analysis Intelligence Index by setting: GPT-6.2 Sol max scores 57, one point above Fable 5.2 medium at 27% of the cost, and no Sol setting reaches Fable high at 62](https://blog.laozhang.ai/posts/en/claude-fable-5-2-vs-gpt-6-2-sol/img/effort-ladder.webp)

"GPT-6.2 Sol Pro" isn't a higher rung on this ladder or a separate model. It's Sol with `reasoning.mode` set to `pro`, an option [OpenAI's reasoning guide](https://developers.openai.com/api/docs/guides/reasoning) describes for hard tasks where higher latency and token use are acceptable, independent of effort. [OpenRouter's GPT-6.2 Sol Pro listing](https://openrouter.ai/openai/gpt-6.2-sol-pro) describes it as "the same underlying model as GPT-6.2 Sol, served with reasoning.mode set to pro", at the same $2 / $10 per token, with a typical request costing several times more. OpenAI's price list has no separate row for it, and the Artificial Analysis rows above are standard mode.

## Where Sol beats Fable on OpenAI's own charts, and where it doesn't

OpenAI's [GPT-6.2 Astra and Sol announcement](https://openai.com/index/introducing-gpt-6-2-astra-sol/) charts Sol and Fable 5.2 on five benchmarks besides the Artificial Analysis index, and they split: Sol ties Fable on Agents' Last Exam, and Fable's top setting leads on the other four, although on BenchCAD Sol at high beats Fable at high. These are vendor-reported numbers on benchmarks OpenAI chose, read from the charts' data labels as of November 20, 2026, with cost per task from the same charts.

| Benchmark | Fable 5.2 at its best | GPT-6.2 Sol at its best | What it shows |
| --- | --- | --- | --- |
| Agents' Last Exam v1.1 | 72.8% at $60.05 (max, the only Fable point) | 73.0% at $1.32 (max) | A tie; Fable's result excludes nine tasks that used a fallback model |
| Terminal-Bench-Science 0.1 | 73.3% at $45.21 (max) | 68.0% at $2.21 (max) | Fable leads at the top; Sol medium's 63.1% beats Fable medium's 60.5% |
| Vals Science Index v1 | 63.7% at $8.79 (one point) | 58.5% at $0.71 | Fable leads by 5.2 points |
| BenchCAD 2.1 Core | 83.5% at $50.77 (max) | 73.1% at $1.60 (max) | Sol high's 64.0% beats Fable high's 61.0%; Fable xhigh and max pull away |
| AutomationBench 1.0.6 | 51.4% at $4.64 (max) | 41.2% at $0.21 (high) | Fable leads by 10.2; both score 40.2% at medium |

The pattern matches the Artificial Analysis ladder. At medium, Sol scores level with or above Fable on all three charts that plot Fable at medium, for a small fraction of the cost per task. At xhigh and max, Fable keeps climbing and Sol flattens: on Terminal-Bench-Science and AutomationBench, Sol's xhigh and max add at most 0.3 points over its high, while Fable gains 3.3 and 7.1 points from high to max. Two caveats from the chart notes apply. Artificial Analysis ran Fable on Terminal-Bench-Science with the mini-swe-agent harness, and Anthropic reports 74.6% for Fable 5.2 in Claude Code at max. And the 0.2-point lead on Agents' Last Exam is a tie in practice, not a win, given that Fable's figure leaves out nine tasks.

## Is GPT-6.2 Sol faster than Fable 5.2? Per call yes, at equal scores no

At the same effort label Sol finishes faster, but when you compare settings that reach the same score, Sol max takes longer than Fable medium. On Artificial Analysis as of November 20, 2026, Sol at high took 5.0 minutes per index task against 15.2 for Fable high, and Sol medium returned its first answer token in 3.6 seconds against Fable medium's 11.2. At equal scores the order flips: Sol max needs 9.6 minutes per task and 83.9 seconds to the first answer token, against 6.5 minutes and 11.2 seconds for Fable medium, which it beats by only one point. Sol xhigh, which ties Fable medium, is slightly quicker per task (5.9 minutes) but waits 34.9 seconds before answering. Artificial Analysis has revised Fable's minutes between reads before, so treat these as of that date.

In the November API tests, Sol at default was faster overall: 3.1x on the 11-answer set and 5.4x on the long log, with a 7.3-second median per call on the hard set against Fable's 25.8. In the coding CLIs the gap mostly disappeared: Sol's median run took 195 seconds and Fable's two runs averaged 221.8, only 1.14x as long. For latency-sensitive work, both sell a faster tier: Fable's fast mode at 2x its standard price, and Sol's Fast at 2x or Ultrafast at 3x until February 28, 2027.

## What rules out Fable 5.2 or GPT-6.2 Sol before quality matters

Check these conditions before running a single quality test, because each one either blocks a model outright or changes its price enough to settle the question. The Fable conditions come from [Anthropic's Fable 5.2 model overview](https://platform.claude.com/docs/en/models/fable-5-2/overview) and its [data retention requirements](https://platform.claude.com/docs/en/manage-claude/api-and-data-retention#model-specific-data-retention-requirements); the Sol conditions from the [GPT-6.2 Sol model page](https://developers.openai.com/api/docs/models/gpt-6.2-sol) and OpenAI's pricing page, as of November 18, 2026.

- If your organization uses zero data retention, Fable 5.2 is out: it requires 30-day retention and returns a 400 error unless your organization is authorized.
- If your pipeline forces a tool call with `tool_choice` set to `any` or a named tool, Fable 5.2 returns 400 until you change it.
- If you work on security or biology research, Fable's safety classifiers route high-risk cybersecurity and most research-biology requests to an older model, so you aren't getting Fable 5.2's answers there anyway.
- If you run high-volume jobs, Fable 5.2 has lower rate limits than Opus 5.5.
- If you send PDFs, Fable 5.2 reads them directly, while Sol takes only text and images, so moving needs a conversion step.
- If single requests pass 272K input tokens, Sol bills the entire request at $4 / $15, and it accepts at most 922,000 input tokens of its 1,050,000-token context; Fable bills its full 1M context at standard rates.
- If your code sends effort `none`, Sol rejects it with a 400 ("Unsupported value: 'none' is not supported with the 'gpt-6.2-sol' model"), as confirmed on November 18, 2026; its lowest setting is `low`.
- If you want Sol in ChatGPT, it's only in ChatGPT Work and Codex, not regular chat.

## Fable 5.2 and GPT-6.2 Sol in ChatGPT and Claude plans

In ChatGPT, GPT-6.2 Sol is only in ChatGPT Work and Codex, not regular chat; in Claude, Fable 5.2 is included in Max, Ultra and premium Team or Enterprise seats up to half of the weekly limit, and billed as usage credits on Pro and standard Team seats. OpenAI's [help article on GPT-6 and other models in ChatGPT](https://help.openai.com/en/articles/20001354-gpt-6-and-other-models-in-chatgpt) says the November 17 rollout "does not add either model to regular Chat"; Plus, Pro, Enterprise and Edu users get Sol in Work and Codex, and Ultrafast Sol needs a Pro plan at $200 or $500. OpenAI doesn't publish message counts for Sol there. On the Claude side, [Anthropic's help article on Fable models on your plan](https://support.claude.com/en/articles/15424964-claude-fable-models-on-your-plan) describes the 50% weekly cap, and the launch-week inclusion for Pro and standard Team ran November 9–16 and has ended. Note that Claude Code runs Fable at medium effort by default, while the Claude API defaults to high, so a Fable comparison in Claude Code starts from a cheaper setting than the API tests above. For API access to Fable without a separate key, see [Claude Fable 5.2 API Key: How to Access It With a Regular Key](https://blog.laozhang.ai/en/posts/claude-fable-5-2-api-key).

## How to test Sol against Fable on your own tasks

Judge the switch by cost per passing result on your own work, measured with the same prompts on both models. Even 10–20 tasks will show whether your work behaves like the November tests, where both models passed everything and Sol cost up to a twelfth as much, or whether some of it only passes on Fable.

1. Pick tasks you already have a pass check for: unit tests, exact answers, or a reviewer's accept or reject. Include the hardest jobs you currently run on Fable at high or above.
2. Run Fable at the effort you use today, set explicitly, since the API defaults to `high` and Claude Code to `medium`. On the Claude API that's `output_config.effort` with adaptive thinking, as described in [Anthropic's effort documentation](https://platform.claude.com/docs/en/build-with-claude/effort).
3. Run Sol at `medium` and `high` on the same prompts, with `reasoning_effort` on Chat Completions or `reasoning.effort` on the Responses API. Use the Responses API if you want `max`, which Chat Completions didn't accept in the November 18 launch test.
4. If you send a long reusable system prompt, put it in a developer message as the first input item on Sol. From GPT-6.2 on, top-level instructions in the Responses API are appended after the input, so they aren't eligible for prompt caching.
5. Record input and output tokens per call and price them yourself, because Fable counts the same prompt as about 1.4–1.5x more tokens and the two bills won't line up by token count.
6. Divide each model's total spend by its number of passes. Re-run any task Sol fails at `xhigh` before assigning it to Fable, and keep on Fable whatever Sol still fails at `max`.

If both models pass the same set, as they did on every November test, the cost per pass decides it, and on those tests Fable cost 6.5–12.2x as much as Sol. Any task that only Fable passes is the work worth paying Fable's price for.

## Sources

External pages this guide links to, in the order they appear. Last updated 2026-11-21.

- [Anthropic's Claude API pricing page](https://platform.claude.com/docs/en/about-claude/pricing) (platform.claude.com)
- [OpenAI's API pricing page](https://developers.openai.com/api/docs/pricing) (developers.openai.com)
- [Artificial Analysis Claude Fable 5.2 vs GPT-6.2 Sol comparison](https://artificialanalysis.ai/models/comparisons/claude-fable-5-2-vs-gpt-6-2-sol) (artificialanalysis.ai)
- [OpenAI's reasoning guide](https://developers.openai.com/api/docs/guides/reasoning) (developers.openai.com)
- [OpenRouter's GPT-6.2 Sol Pro listing](https://openrouter.ai/openai/gpt-6.2-sol-pro) (openrouter.ai)
- [GPT-6.2 Astra and Sol announcement](https://openai.com/index/introducing-gpt-6-2-astra-sol/) (openai.com)
- [Anthropic's Fable 5.2 model overview](https://platform.claude.com/docs/en/models/fable-5-2/overview) (platform.claude.com)
- [data retention requirements](https://platform.claude.com/docs/en/manage-claude/api-and-data-retention) (platform.claude.com)
- [GPT-6.2 Sol model page](https://developers.openai.com/api/docs/models/gpt-6.2-sol) (developers.openai.com)
- [help article on GPT-6 and other models in ChatGPT](https://help.openai.com/en/articles/20001354-gpt-6-and-other-models-in-chatgpt) (help.openai.com)
- [Anthropic's help article on Fable models on your plan](https://support.claude.com/en/articles/15424964-claude-fable-models-on-your-plan) (support.claude.com)
- [Anthropic's effort documentation](https://platform.claude.com/docs/en/build-with-claude/effort) (platform.claude.com)
