# Gemini 4 Argon vs Claude Fable 5.1: A Tie on Score, Not on Cost

> Gemini 4 Argon and Claude Fable 5.1 tie on Artificial Analysis, yet Fable cost 8.6x Argon on everyday tasks in an October–November test, 4.3x from December 1.

- URL: https://blog.laozhang.ai/en/posts/gemini-4-argon-vs-claude-fable-5-1
- Published: 2026-11-04
- Updated: 2026-11-04
- Author: LaoZhang AI Team (https://blog.laozhang.ai/en/about)
- Topic: Model Comparisons
- Tags: Gemini 4 Argon, Claude Fable 5.1, AI Model Comparison, API Pricing, Benchmarks, Coding Agents

---
For most API work in November 2026, start with Gemini 4 Argon (`gemini-4-argon`) at its default thinking level, and keep Claude Fable 5.1 (`claude-fable-5-1`) for the jobs where it measurably leads: office deliverables such as reports and spreadsheets, the hardest reasoning questions, and code generation of the kind FrontierCode measures. On quality the two are tied as of November 4, 2026: Argon (High) scores 50.41 and Fable 5.1 (xhigh) 50.02 on the [Artificial Analysis Intelligence Index v5.0.1](https://artificialanalysis.ai/models/comparisons/gemini-4-argon-vs-claude-fable-5-1), and Fable 166.36 against Argon 166.07 on [Epoch AI's Capabilities Index](https://epoch.ai/benchmarks/eci). Both gaps are inside the noise.

Cost is where they split. On five everyday tasks run through the same API gateway (Argon on November 4, 2026, Fable on October 26, with its setup rechecked on November 4), Fable 5.1 at its default setting cost 8.6 times as much as Argon at its default, and both passed every hidden test. Argon's introductory price of $2 input and $10 output per million tokens ends on November 30. From December 1 it doubles to $4 and $20, and on the same tasks Fable's bill drops to 4.3 times Argon's.

Two choices reverse that picture. If you turn Argon's thinking up to `high`, it costs more than Fable's default on harder tasks once the December price applies. And if you are really choosing within Anthropic's lineup, Anthropic itself says to start with Claude Opus 5.5 and move to Fable only when Opus at higher effort falls short.

## How much better is Fable 5.1 than Gemini 4 Argon? A tie on AA v5.0.1

Neither model is better on the overall index. Artificial Analysis (AA) tests Argon only at its High thinking level, where it scores 50.41, and Fable 5.1 tops out at 50.02 with `xhigh` effort. Differences under about one point are within the index's run-to-run noise, so 50.41 against 50.02 is a tie, not a win. The figures below are from AA's [model comparison page](https://artificialanalysis.ai/models/comparisons/gemini-4-argon-vs-claude-fable-5-1) and its [Claude Fable 5.1 page](https://artificialanalysis.ai/models/claude-fable-5-1), as of November 4, 2026.

| Model and setting (AA v5.0.1) | Index score | Cost per index task | Time per task | Time to first token |
| --- | ---: | ---: | ---: | ---: |
| Gemini 4 Argon, High | 50.41 | $7.68 (at $4/$20) | 1,362.9 s | 41.5 s |
| Claude Fable 5.1, max | 49.16 | $12.99 | 1,066.4 s | 252.5 s |
| Claude Fable 5.1, xhigh | 50.02 | $10.85 | 965.7 s | 79.5 s |
| Claude Fable 5.1, high (API default) | 47.17 | $7.54 | 719.8 s | 18.3 s |
| Claude Fable 5.1, medium | 43.17 | $5.56 | 537.1 s | 14.7 s |
| Claude Fable 5.1, low | 38.68 | $4.47 | 421.3 s | 2.6 s |

The effort names do not line up across vendors, so compare at matched cost or matched score instead. Google's thinking levels for Argon are `low`, `medium` and `high`, with `medium` as the API default, so AA's High run is one step above what you get without setting anything. Fable's effort runs from `low` to `max` with `high` as the default. At matched cost, Fable at `high` ($7.54 per task) is about the same price as Argon High ($7.68) and scores 3.2 points lower. At a matched score, Fable at `xhigh` costs $10.85, about 1.4 times Argon. Fable at `max` scores below `xhigh` on this index while costing more, so `max` buys nothing here.

AA already prices Argon at its December rate. Through November 30 every Argon rate is half that, so the same index task works out to roughly $3.84, a simple halving of AA's figure rather than a number AA publishes. Against that, Fable at `high` costs about 2.0 times as much and Fable at `xhigh` about 2.8 times.

![Scatter chart of AA Index v5.0.1 score against cost per task: Argon High at 50.41 for $7.68, and Fable 5.1 from low to max, with high at 47.17 for $7.54 and xhigh at 50.02 for $10.85](https://blog.laozhang.ai/posts/en/gemini-4-argon-vs-claude-fable-5-1/img/aa-score-vs-cost.webp)

If you have seen the two described as "0.79 index points apart, at 5x the price", or Argon placed "around 53, on par with Fable", those numbers come from AA's v4.3.2 index, which AA used in September and October. AA moved to v5.0 and then, in November, to v5.0.1 (a regrade of the CritPt evaluation, with the same nine evaluations and weights), and scores from different versions cannot be compared. The same goes for older Epoch figures of 165 for both models: the November 3 update puts them at 166.07 and 166.36 with overlapping confidence intervals. And if you read that Fable is the better choice "because you can actually call it", that dates from before Argon became generally available on October 13, 2026. On the [Vals Index](https://www.vals.ai/benchmarks/vals_index), now version 2.2, Argon scores 69.24% and ranks 4th of 32, while Fable 5.1 does not appear on the current leaderboard at all, so the older Vals pairing of 68.90% against 65.83% no longer stands.

## Where Argon leads and where Fable leads, benchmark by benchmark

The tie on the index hides clear splits by kind of work. On AA's own runs, Argon leads the agentic and terminal evaluations, while Fable leads office-style knowledge work, FrontierCode and the hardest reasoning questions. The table shows AA v5.0.1 per-evaluation results as of November 4, 2026, with Fable at both of its top effort levels.

| Evaluation (AA v5.0.1) | Argon High | Fable 5.1 max | Fable 5.1 xhigh | Ahead |
| --- | ---: | ---: | ---: | --- |
| AutomationBench-AA | 77.5% | 59.4% | 57.8% | Argon |
| Terminal-Bench 4.0 | 57.1% | 52.0% | 55.1% | Argon |
| Terminal-Bench-Science | 52.9% | 43.3% | 40.5% | Argon |
| AA-Briefcase (office and knowledge work) | 1484 | 1669 | 1649 | Fable |
| GDP.pdf | 21.8% | 26.2% | 26.2% | Fable |
| FrontierCode 2.0 | 25.7% | 31.6% | 35.7% | Fable |
| HLE-Diamond | 56.7% | 60.1% | 58.8% | Fable |
| AA-Omniscience | 46.1 | 46.3 | 44.3 | Tie |
| CritPt v2 Hard | 52.5% | 52.5% | 60.0% | Tie with max, Fable xhigh ahead |

Running the whole index cost AA $6,103 for Argon and $13,327 for Fable 5.1 at `max`. The pattern matters more than any single row: if your work is a model calling tools, driving a terminal or automating a workflow, Argon is ahead by margins that are not noise; if it is producing documents, answering expert-level questions or writing code that has to be correct the first time, Fable is.

Google's own [Gemini 4 Argon launch post](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/) from September 30, 2026 tells a stronger story for Argon, but Google ran both models itself, with Argon at its highest thinking level, and nobody has independently reproduced the table. In it, Argon is ahead on 15 of the 17 benchmarks it shares with Fable 5.1, with the widest gaps on AutomationBench (51.3 against 31.4), Harvey Legal Agent (19.6 against 6.7), LABBench 2 (88.8 against 68.6), Chartography (71.6 against 46.2) and long-context GraphWalks at 256K to 1M tokens (84.2 against 65.0). Fable is ahead on FrontierSWE v2 (56.3 against 55.0) and on Terminal-bench 4.0 (57.9 against 57.4), a result that flips on AA's independent run. A Google chart reported by VentureBeat also put Argon's indirect prompt-injection attack success rate on Gray Swan's test at 0.7%, against 1.0% for Fable 5.1.

## What one task costs on each, now and after December 1

Fable 5.1's price per token is five times Argon's through November 30, 2026, and two and a half times from December 1. The rates below are from [Google's Gemini API pricing page](https://aistudio.google.com/docs/pricing) and [Anthropic's Claude API pricing page](https://platform.claude.com/docs/en/about-claude/pricing) as of November 4, 2026, in US dollars per million tokens.

| Rate per 1M tokens | Gemini 4 Argon to Nov 30 | Gemini 4 Argon from Dec 1 | Claude Fable 5.1 |
| --- | ---: | ---: | ---: |
| Input | $2.00 | $4.00 | $10.00 |
| Output, thinking included | $10.00 | $20.00 | $50.00 |
| Cached input read | $0.10 | $0.20 | $0.25 |
| Cache write | implicit caching only | implicit caching only | $12.50 (5 min), $20 (1 hour) |
| Batch input and output | $1 and $5 | $2 and $10 | $5 and $25 |

Neither model charges more for long prompts: Argon has no separate tier above 200K tokens, and Fable bills its full 1M window at the standard rate. Fable's rates are multiplied by 1.1 when you pin processing to the US with `inference_geo: "us"`. Argon has no free API tier, and Google has not said whether the introductory price will be extended.

The bill for the same work is the price ratio multiplied by the token ratio, worked out separately for input and output. The token ratio is not 1, because Fable 5.1 counts the same text as more tokens: in a hands-on count through the same gateway (Argon on November 4, 2026, Fable on October 26, 2026, identical text), Fable reported 1.45 times Argon's tokens for a short English passage and 1.38 times for a 6,723-word English article (15,335 against 11,110). So uncached English input costs about 7.3 times as much on Fable now (5 × 1.45) and 3.6 times from December 1 (2.5 × 1.45); for the long article it is 6.9 times and 3.5 times. Anthropic says Fable uses the tokenizer introduced with Opus 4.7, which produces [roughly 30% more tokens than older Claude models](https://platform.claude.com/docs/en/build-with-claude/token-counting); neither vendor publishes a Claude-to-Gemini comparison.

Prompts in other languages widen the gap. The same count measured Fable at 1.72 times Argon's tokens in Russian, 1.78 in Chinese, 1.85 in Japanese, 1.97 in Spanish and 2.26 in Korean, one passage per language, which puts the uncached input bill for Korean text at 11.3 times Argon's now and 5.7 times from December 1. If most of your input is not English, measure your own text before relying on the English ratio.

The output side depends on how much each model thinks, and thinking tokens are billed as output on both. In the eight-task test (Argon run on November 4, 2026, Fable on October 26), Fable produced 1.4 times as many output tokens as Argon at its default, but only a quarter as many as Argon at `high`, which is why Argon's thinking level can swing the result more than the price list does.

Caching narrows the input gap if your prompts repeat a long prefix. Fable's $0.25 cache reads apply only to content you mark with `cache_control`, and each write costs 1.25 times input for five minutes or 2 times for an hour. Argon caches implicitly, with no API to control it, once a prompt passes about 4,096 tokens according to Google's Gemini API docs (some Google Cloud pages say 6,144). Per cached token, Fable costs 2.5 times Argon now and 1.25 times from December 1. The mechanics are in [Claude Prompt Caching: How It Works, Costs, 512-Token Minimum](https://blog.laozhang.ai/en/posts/claude-prompt-caching) and [Gemini Context Caching: Implicit vs Explicit, Pricing, Hit Rates](https://blog.laozhang.ai/en/posts/gemini-context-caching), and full rate cards in [Claude Fable 5.1 API Pricing: What Changed and What You'll Pay](https://blog.laozhang.ai/en/posts/claude-fable-5-1-api-pricing) and [Gemini API Pricing 2026: Token Rates, Free Tier, and Key Cost](https://blog.laozhang.ai/en/posts/gemini-api-pricing).

## Same eight tasks, same gateway: every answer passed, the cost did not match

In a hands-on test, Gemini 4 Argon (run on November 4, 2026) and Claude Fable 5.1 (run on October 26, 2026) answered the same eight tasks, each graded by hidden pytest tests, and all 33 graded answers passed. The tasks did not separate the models on quality, so what the test shows is cost, tokens and time at an equal pass rate.

| Configuration | 5 everyday tasks, to Nov 30 / from Dec 1 | 3 harder tasks × 2, to Nov 30 / from Dec 1 | Output tokens, all 11 calls | Wall time, all 11 calls |
| --- | --- | --- | ---: | ---: |
| Argon, default (no thinking setting) | $0.0384 / $0.0768 | $0.0556 / $0.1112 | 6,859 (1,531 thinking) | 128 s |
| Argon, thinking `high` | $0.1364 / $0.2728 | $0.2565 / $0.5130 | 36,746 (31,016 thinking) | 532 s |
| Fable 5.1, default (effort `high`) | $0.3310 | $0.3073 | 9,327 (thinking included) | 116 s |

Against Argon at its default, Fable cost 8.6 times as much on the everyday round through November 30 and 4.3 times from December 1; on the harder round the gap was 5.5 times, then 2.8 times. Against Argon at `high`, Fable cost 2.4 times and then 1.2 times on the everyday round, and on the harder round 1.2 times now but 0.60 times from December 1. In other words, once Argon's price doubles, Argon at `high` thinking costs more than Fable at its default on these harder tasks. Speed followed thinking too: Argon at its default took about as long as Fable (128 against 116 seconds for all 11 calls), and Argon at `high` was about 4.6 times slower.

![Bar chart of Fable 5.1's bill as a multiple of Argon's in the eight-task test, before and after December 1: 8.6x to 4.3x on everyday tasks against Argon's default, and 1.2x to 0.60x on harder tasks against Argon at high thinking](https://blog.laozhang.ai/posts/en/gemini-4-argon-vs-claude-fable-5-1/img/eight-task-cost-ratio.webp)

The everyday round was a word problem, merging intervals, parsing durations, counting top words and summarizing a 6,723-word article, one sample each. The harder round was an expression evaluator, a CSV line parser and a digit-sum puzzle, two samples each. Both models ran through the LaoZhang API gateway, which lists `gemini-4-argon` at $2 / $10 and `claude-fable-5-1` at $10 / $50 on its [pricing page](https://api.laozhang.ai/api/pricing), the same as Google and Anthropic, and has not published an Argon price for December. Argon used the Gemini-native `generateContent` endpoint without streaming and a 32,000-token output limit; Fable used the Anthropic Messages API with streaming. No caching and no tools were involved, and costs are list-price arithmetic from the token counts each API returned. One fresh Fable call on November 4 returned the same model ID and the same input tokens as on October 26, with 674 output tokens against 773, ordinary variation from run to run.

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

The limits are as important as the numbers. These are small, well-specified tasks with one or two samples each, no agent loops and no long outputs, so the test says nothing about which model is better; AA's index and Google's launch table remain the quality reference. Argon at its default reported zero thinking tokens on 9 of 11 calls and thought only on the digit-sum puzzle, so on harder, multi-step agent work it will likely think more, and cost more, than it did here.

## Which to use for which work in November 2026

Pick by the kind of work first, then by the date and the thinking level you plan to run. The starting points below combine the AA per-evaluation results, the price periods and the site test; each one is a first candidate to test, not a guaranteed winner.

| Kind of work | Start with | Why | What would change it |
| --- | --- | --- | --- |
| Agents, tool calling, terminal and workflow automation | Argon, default thinking | Leads AutomationBench-AA, Terminal-Bench 4.0 and Terminal-Bench-Science; a fraction of Fable's bill | Your loop only passes at `high` thinking, which from December 1 can cost more than Fable's default |
| Long documents or large codebases in the prompt | Argon | Same 1M-token context, lower input price, fewer tokens for the same English text | A long prefix reused many times, where Fable's marked cache reads close the gap |
| Reports, documents, spreadsheets and slides | Fable 5.1 | Leads AA-Briefcase and GDP.pdf | Argon's drafts pass your own review at a fraction of the cost |
| Expert-level reasoning and research questions | Fable 5.1 at `xhigh` | Leads HLE-Diamond; `xhigh` beats `max` on the index | The cost: $10.85 per AA task against $7.68 for Argon High at the December price |
| Code that must be right first time (FrontierCode-style) | Fable 5.1 at `xhigh` | 35.7% against 25.7% on FrontierCode 2.0 | Repository and terminal agent work, where Argon leads on AA and in Google's table |
| One answer longer than 65,536 tokens | Fable 5.1 (128K output) | Argon caps a synchronous request at 65,536 tokens | Argon's long decode mode, which goes up to 1,048,576 tokens |
| Interactive use where the first token matters | Fable 5.1 at `high` or lower | AA measures 18.3 s to first token at `high`, against 41.5 s for Argon High | At default settings the site test saw similar total times |

Two practical limits sit behind the long-output row. Google's [long decode docs](https://aistudio.google.com/docs/long-decode) cap a single non-streaming request at 32,768 output tokens and require `background=true` or client-side continuation to go past 65,536; [Gemini 4 Argon API finish_reason: CONTINUATION and Empty Replies](https://blog.laozhang.ai/en/posts/gemini-4-argon-api-finish-reason) covers how that looks in practice. Through the LaoZhang gateway, a single Claude response is capped at 32,768 output tokens, so Fable's 128K advantage applies only on Anthropic's API and the clouds that host it. Fable 5.1 also requires 30-day data retention on Anthropic's API, which rules it out if your agreement forbids retention.

If you already work on Claude, Anthropic's [models overview](https://platform.claude.com/docs/en/about-claude/models/overview) is direct about the order: "If you're unsure which model to use, start with Claude Opus 5.5 for most workloads. Use Claude Fable 5.1 for demanding reasoning and long-horizon agentic work, or when your evals on Claude Opus 5.5 at higher effort still fall short." The trade-off is in [Claude Fable 5.1 vs Opus 5.5: Which Should You Use?](https://blog.laozhang.ai/en/posts/claude-fable-5-1-vs-opus-5-5), and Argon against Opus is in [Gemini 4 Argon vs Claude Opus 5.5 vs GPT-6.1 Sol: Cost and Access](https://blog.laozhang.ai/en/posts/gemini-4-argon-vs-claude-opus-5-5-vs-gpt-6-1-sol).

Three other names come up in the same decision. OpenAI's flagship against Fable is covered in [GPT-6 Astra vs Claude Fable 5.1: Same Price, 1.5x the Tokens](https://blog.laozhang.ai/en/posts/gpt-5-6-sol-vs-fable-5-vs-opus-4-8). There is no Fable 5.5 in Anthropic's docs or release notes as of November 4, 2026, and Fable 5.1 is still marked "Latest"; [Fable 5.5 vs Gemini 4 vs GPT-6.1: Which Can You Use Today?](https://blog.laozhang.ai/en/posts/gemini-4-vs-gpt-6-vs-fable-5-2) traces the rumor. Gemini 4 Carbon, announced October 28, is open only to a group of cyber defenders with no public model ID or date, and Google has not said Argon will be retired; [Gemini 4 Carbon vs Gemini 4 Argon: Slower Start, Faster Agents](https://blog.laozhang.ai/en/posts/gemini-4-carbon-vs-gemini-4-argon) covers what is known.

## Gemini app, Claude app, Antigravity and Claude Code: where each one runs

Both models run outside the API too, but only on paid plans, and no consumer plan covers API calls. In the Gemini app, Argon has been available since October 13, 2026 on Google AI Pro ($19.99 a month in the US) and Google AI Ultra ($99.99 or $199.99), which gets the highest access; the $4.99 Google AI Plus plan includes only the Flash model, according to [Google's Gemini subscriptions page](https://gemini.google/us/subscriptions/?hl=en). The same Pro and Ultra plans give Argon quota in the AI Studio web interface. For coding, as of late October, Argon runs in Google's Antigravity with a separate weekly quota on paid Pro and Ultra, while the Gemini CLI documentation says Argon is not optimized there and points to Antigravity.

On the Claude side, Fable 5.1 is paid with usage credits from the first message on Pro ($20 a month), and Max ($100) and Ultra ($200) include Fable up to half of the plan's weekly limit, after which you pay with usage credits or switch models, per [Anthropic's help article on Fable models](https://support.claude.com/en/articles/15424964-claude-fable-models-on-your-plan). Max and Ultra are the October 30, 2026 names for what used to be Max 5x and Max 20x, and they include $100 and $200 of monthly API credits. Claude Code needs version 2.1.255 or later to use Fable 5.1. How far each plan stretches is covered in [Gemini 4 Argon Usage Limits: Pro vs Ultra, Resets, API Credits](https://blog.laozhang.ai/en/posts/gemini-4-argon-usage-limits); Google publishes no prompt counts for Argon.

## How to test Argon and Fable 5.1 on your own prompts before switching

A few dozen of your own prompts, graded the same way and priced at both of Argon's price periods, will tell you more than any index, because both the quality split and the token count depend on your kind of work and your language. Run the comparison like this:

1. Collect 20 to 50 real prompts from the work you plan to move, each with a check you can run without looking at which model answered: unit tests, a reference answer or a written rubric.
2. Call Argon through the Gemini API with `thinkingConfig.thinkingLevel` set explicitly, once at `medium` (the default) and once at `high`, and set the output limit to at least 32,768 tokens as Google recommends. Some OpenAI-compatible gateways, LaoZhang's among them, ignore `reasoning_effort` for Argon, so use the native field.
3. Call Fable 5.1 at effort `high` (the default) and at `xhigh`, with a large `max_tokens`, as [Anthropic's effort guide](https://platform.claude.com/docs/en/build-with-claude/effort) advises for `high` and above.
4. Run every prompt at least twice per configuration, because output length and thinking vary from run to run.
5. Log the input, output and cached token counts each API returns, then price every Argon run twice, at the November rate and at the December rate, and every Fable run at its single rate.
6. Divide each configuration's total cost by the number of answers that passed. That cost per accepted answer, before and after December 1, is the number to decide on.

If Argon at `medium` passes the checks, the cost case for it is strong in both price periods. If it only passes at `high`, compare that configuration against Fable at `high` with the December prices, since that is where the site test saw the gap close and then flip.

## Gemini 4 Argon vs Claude Fable 5.1: other questions

### Does Batch or Flex processing change the gap between Argon and Fable 5.1?

Not the per-token ratio. As of November 4, 2026, Argon's Batch and Flex rates are $1 input and $5 output per million tokens through November 30 and $2 and $10 from December 1, while Fable 5.1's Batch rate is $5 and $25, so Fable stays five times Argon's price per token and then two and a half times. The difference in token counts still applies on top. Argon's Priority tier runs $3.60 / $18, rising to $7.20 / $36 in December.

### Is "Gemini 4 Pro" the same model as Gemini 4 Argon?

"Gemini 4 Pro" was a name that circulated before launch; Google's model is Gemini 4 Argon, with the API ID `gemini-4-argon`, generally available since October 13, 2026. "Gemini 4" on its own can also mean Gemini 4 Carbon, which is not publicly available, so check the full name before comparing numbers.

### Can you try Gemini 4 Argon or Claude Fable 5.1 for free?

Not through the official services as of November 4, 2026. The Gemini API has no free tier for Argon, the free AI Studio web plan gets no Argon quota, and the free Claude plan does not include Fable 5.1. Paying per token through the API is the simplest way to try both on real work: in the eight-task test, all eleven graded calls came to $0.094 on Argon at its default (run November 4, 2026) and $0.638 on Fable 5.1 (run October 26, 2026) at list price.

## Sources

External pages this guide links to, in the order they appear. Last updated 2026-11-04.

- [Artificial Analysis Intelligence Index v5.0.1](https://artificialanalysis.ai/models/comparisons/gemini-4-argon-vs-claude-fable-5-1) (artificialanalysis.ai)
- [Epoch AI's Capabilities Index](https://epoch.ai/benchmarks/eci) (epoch.ai)
- [Claude Fable 5.1 page](https://artificialanalysis.ai/models/claude-fable-5-1) (artificialanalysis.ai)
- [Vals Index](https://www.vals.ai/benchmarks/vals_index) (vals.ai)
- [Gemini 4 Argon launch post](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/) (blog.google)
- [Google's Gemini API pricing page](https://aistudio.google.com/docs/pricing) (aistudio.google.com)
- [Anthropic's Claude API pricing page](https://platform.claude.com/docs/en/about-claude/pricing) (platform.claude.com)
- [roughly 30% more tokens than older Claude models](https://platform.claude.com/docs/en/build-with-claude/token-counting) (platform.claude.com)
- [long decode docs](https://aistudio.google.com/docs/long-decode) (aistudio.google.com)
- [models overview](https://platform.claude.com/docs/en/about-claude/models/overview) (platform.claude.com)
- [Google's Gemini subscriptions page](https://gemini.google/us/subscriptions/?hl=en) (gemini.google)
- [Anthropic's help article on Fable models](https://support.claude.com/en/articles/15424964-claude-fable-models-on-your-plan) (support.claude.com)
- [Anthropic's effort guide](https://platform.claude.com/docs/en/build-with-claude/effort) (platform.claude.com)