Claude Fable 5.2 vs GPT-6.1 Terra

By Versus Tools AI Team · Published · Updated · 9 min read · AI Models & News

We may earn a commission if you buy through links on this page. Read the affiliate disclosure · How we review tools

Head to head

Claude Fable 5.2$8
GPT-6.1 Terra$1
Fable for quality. Terra for high volume.
Claude Fable 5.24.4
GPT-6.1 Terra4.4

Quick answer. Claude Fable 5.2 is the stronger published pick for coding, agent, and reasoning quality. GPT-6.1 Terra's standard list price is $1 and $5 per million tokens, 8 times Fable's $8 and $40, and Artificial Analysis records a shorter time per task. This page uses official docs, pricing pages, and public reports.

Tools compared in this article
ToolRatingBest forPriceWebsite
Claude Fable 5.24.4/5Coding and hard reasoning$8 / $40 per 1MVisit Claude Fable 5.2 →
GPT-6.1 Terra4.4/5High-volume API work$1 / $5 per 1MVisit GPT-6.1 Terra →

Claude Fable 5.2 vs GPT-6.1 Terra is a quality-versus-volume choice. Fable, released Nov 9, 2026, is the stronger model on the coding, agent, and reasoning scores its maker and Artificial Analysis have published. Terra, dated Oct 26, 2026 on OpenAI’s research index, is the high-volume model: the standard list price is $1 per million input tokens and $5 per million output tokens, which is 8 times lower than Fable’s $8 and $40, and Artificial Analysis records a shorter time per task.

This page is based on official docs, pricing pages, and public reports, checked Nov 17, 2026. The 4.4 marks are tied on purpose. Each mark means the model has a documented job on this page. Neither mark is an Intelligence Index score.

Anthropic’s own benchmark table, Mythos 5.2, and the cache card live on Claude Fable 5.2. The month’s other launches are on the AI model release timeline and in best AI models.

How this comparison was made

We read Anthropic’s Nov 9, 2026 announcement, the Fable 5.2 model overview, and the pricing facts already sourced on Claude Fable 5.2. For Terra we read OpenAI’s launch post, the model page at developers.openai.com for gpt-6.1-terra, and Artificial Analysis’s comparison page, re-read Nov 17, 2026. The Oct 26 article on Artificial Analysis is used for the claims it prints about GPT-5.6 Terra. BenchLM’s comparison page was opened the same day. Its overall point estimates are noted below and are not used as the ranking, because that page says its ranges do not establish rank confidence and many cells are still incomplete.

A lower token price is not a quality ranking. A higher Intelligence Index is not an invoice.

Side-by-side specs

Claude Fable 5.2GPT-6.1 Terra
ReleasedNov 9, 2026Oct 26, 2026, on OpenAI’s research release index. The launch post says “starting today” and does not print a calendar date.
MakerAnthropic. Mythos 5.2 is the same underlying model, limited to verification programs.OpenAI
API idclaude-fable-5-2gpt-6.1-terra
Context1 million tokens1,050,000 on the model page. 922,000 max input. Artificial Analysis prints 1M.
Max output128K tokens128,000 tokens
Input and outputText and images in. Text out.Text and image in. Text out. Audio and video are listed as unsupported.
Knowledge cutoffJun 2026, on the model overviewJun 01, 2026, on the model page
Standard price$8 / 1M input, $40 / 1M output$1 / 1M input, $5 / 1M output
Cached inputCache read $0.20$0.05
Cache write5-minute write $10. 1-hour write $16.$1.25 for 30-minute retention. $2.50 for 12-hour retention.
DiscountBatch API $4 / $20Batch and Flex are 50% lower than Standard
Long promptsNot restated here. See the Fable overview.Above 272K input tokens, the model page prices the full request at 2x input and cache rates and 1.5x output.

OpenAI’s model page also says fast mode is 2x Standard, and that regional processing adds a 10% premium where it is available. Default reasoning effort on that page is medium, with none, low, high, xhigh, and max also listed. Anthropic’s API default for Fable is a separate setting, covered on Claude Fable 5.2.

Price

Standard rates per million tokens. The 8 times gap is this table, input against input and output against output. It is not a blended cost per task.

Fable 5.2Terra
Input$8$1
Output$40$5
Cached input$0.20 cache read$0.05
Cache write$10 for 5 minutes, $16 for 1 hour$1.25 for 30 minutes, $2.50 for 12 hours
Batch$4 / $2050% below Standard. Flex is also 50% below Standard.

Artificial Analysis, on the comparison page read Nov 17, 2026, prints the same $8 / $40 and $1 / $5 pair at every effort row, cache hit $0.20 and $0.05, and cache write $10 and $1.25. The $1.25 Terra cache write now also appears on OpenAI’s model page, for 30-minute retention. The launch post still prints only the $0.05 cached-input rate.

Its cost per Intelligence Index task:

EffortFableTerra
Low$3.10$0.0713
Max$26.54$0.96

That low-effort pair is about 43 times the dollars per task, not 8 times. The models are spending different amounts of tokens. Use the list price to compare a token. Use this column only as Artificial Analysis’s weighted task cost.

The Oct 26 Artificial Analysis article says Terra’s token price is half or less of GPT-5.6 Terra’s, at $1 / $5 against $2 / $12, with cached input at $0.05 against $0.20. Those are predecessor prices, not Fable prices.

Quality, from published scores

Every number in this section belongs to the publisher named on the row. Nothing here was rerun for this page.

OpenAI, on the Terra launch post, at maximum reasoning: 50.3% on Terminal-Bench 4.0, against 21.5% for GPT-5.6 Terra. OpenAI says Sol and Astra remain the better choice for the most complex coding tasks. On OpenAI’s internal benchmark of error-inducing prompts, the error rate at low effort fell from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra. At maximum effort, OpenAI says the error rate fell by 18 percentage points to 4.3%.

Anthropic’s Nov 9 table is the one copied on Claude Fable 5.2. The rival column in that table is GPT-6 Astra, not GPT-6.1 Terra. Do not read Astra’s cells as Terra’s.

Artificial Analysis Intelligence Index v5.0.1, read Nov 17, 2026:

Fable 5.2Terra
Intelligence Index, max6551
Intelligence Index, high6242
Intelligence Index, low5429
Terminal-Bench 4.0, max59.6%55.6%
Terminal-Bench 4.0, xhigh61.6%46.5%
Terminal-Bench 4.0, low54.5%15.2%

On that page, Fable’s xhigh Terminal-Bench 4.0 score is higher than its max score. Max effort is not the top cell in that row. Terra’s max score, 55.6%, is above OpenAI’s own 50.3%. They are different runs. This page does not average them.

Time per task and output speed, same page:

FableTerra
Time per task, low5.4 minutes1.2 minutes
Time per task, max40.8 minutes14.0 minutes
Output speed, low111 tokens per second64 tokens per second
Output speed, max112 tokens per second71 tokens per second

Terra finishes the indexed task sooner in these rows. Fable emits output tokens faster. “Faster” on this page means the time-per-task column, plus OpenAI’s description of Terra as a fast model against Astra and Sol. It does not mean a higher token-per-second rate than Fable.

A weekly note at ashutoshai.com, for Nov 9 to 13, 2026, also says Fable took first on Artificial Analysis v5 at 65. That matches the max cell above. It is a secondary summary of the index, not a second measurement.

BenchLM’s comparison, updated Nov 16, 2026, prints an overall point estimate of 93.84 for Fable and 78.73 for Terra, and it says those ranges do not establish rank confidence. Several of its category cells are marked estimated or still coming. This page does not use that overall estimate as a winner.

Claude Fable 5.2

👍 Pros

  • Artificial Analysis Intelligence Index 65 at max, against 51 for Terra
  • Anthropic's Nov 9 table, on the Fable page, is the coding and agent set for this model
  • 1 million-token context, 128K max output, cache reads at $0.20

👎 Cons

  • $8 and $40 is 8 times Terra's standard input and output rates
  • Artificial Analysis's low-effort task cost is $3.10 against $0.0713
  • Mythos, not Fable, is the version for the cyber and biology work Fable's classifiers block

GPT-6.1 Terra

👍 Pros

  • $1 and $5 per million tokens, with batch and Flex at 50% of Standard
  • Shorter Artificial Analysis time per task at every effort row we copied
  • OpenAI's published step up from GPT-5.6 Terra on Terminal-Bench 4.0, 50.3% against 21.5%

👎 Cons

  • The same index puts max effort at 51 against Fable's 65
  • OpenAI still points the hardest coding work at Sol and Astra
  • Prompts over 272K input tokens are billed at 2x input and 1.5x output for the whole request

Who should use which

Use Claude Fable 5.2 when the task is the work Anthropic aimed the Nov 9 model at, and when the published quality tables are the constraint you care about: long agent sessions, hard coding, and reasoning where the Intelligence Index gap shows up. Anthropic’s own routing note, repeated on the Fable page, is to start most coding and knowledge work on Opus 5.5, at $4 and $20, and move up to Fable when the task needs it. The spec, the safeguard split with Mythos, and Anthropic’s full table are on Claude Fable 5.2.

Use GPT-6.1 Terra when you are paying for volume. The standard token is 8 times cheaper on input and on output, batch and Flex take another 50% off, and the Artificial Analysis time-per-task column is the shorter one. OpenAI’s launch framing is the same job: invoices, forms, visual checks, and other work you would rather run many times. For the most complex coding tasks, use the model OpenAI names, Sol or Astra, or use Fable. Terra’s 50.3% is a large step up from GPT-5.6 Terra. It is not the top cell on the Artificial Analysis Terminal-Bench row against Fable.

Use the list price and the index for different decisions. Eight times cheaper per token does not mean eight times cheaper per finished task. The low-effort cost column is the illustration. Re-check both vendors on the day you pay.

How to access them

Fable, from Anthropic’s model overview: the Claude API and Claude Platform on AWS as claude-fable-5-2, Amazon Bedrock as global.anthropic.claude-fable-5-2, Claude on Google Cloud, and Microsoft Foundry. The announcement also covers eligible paid Claude plans. The pages read for Claude Fable 5.2 do not name a mainland China endpoint.

Terra, from the launch post and the model page: ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu, and the API as gpt-6.1-terra. Amazon Bedrock’s id on the model page is openai.gpt-6.1-terra. Google Cloud is listed as a private preview. Batch is supported. Chat Completions and the Responses API are supported. The supported-countries article, as read for this site’s GPT Image 2.6 vs Seedream 5.5 page on Nov 17, 2026, does not list mainland China. This comparison does not add an unofficial route for either lab.

Ready-made AI writing apps

Fable and Terra are models you call. If the job is marketing copy in a writing product, rather than an API bill, Writesonic is in this site’s catalog. Cursor is the catalog editor when the work is code around a model. Neither product is a benchmark of Fable or Terra. A commission does not change the $8 / $40 or $1 / $5 rates.

Sources

Read on Nov 17, 2026, unless a page’s own date is noted.

Pricing last checked Nov 17, 2026. The 4.4 marks are not a hands-on score. This page contains affiliate links. A commission does not change which model is recommended for which job.

Frequently asked questions

Which is better, Claude Fable 5.2 or GPT-6.1 Terra?

For coding, agent, and reasoning quality, the published scores favor Claude Fable 5.2. Artificial Analysis Intelligence Index v5.0.1, read Nov 17, 2026, is 65 for Fable at max and 51 for Terra at max. OpenAI's own Terminal-Bench 4.0 figure for Terra is 50.3 percent at maximum reasoning, and OpenAI says GPT-6.1 Sol and GPT-6 Astra remain the better choice for the most complex coding tasks. For a high-volume bill, Terra's standard list price is $1 and $5 per million tokens against Fable's $8 and $40.

How much cheaper is GPT-6.1 Terra per token?

On the standard list rates, Terra is 8 times cheaper on input and on output: $1 against $8, and $5 against $40, per million tokens. Cached input is $0.05 against Fable's $0.20 cache read, which is 4 times, not 8. A task can still cost a different multiple, because the models do not use the same number of tokens. Artificial Analysis's cost per Intelligence Index task at the low setting is $0.0713 for Terra and $3.10 for Fable.

Which is faster?

On the Artificial Analysis comparison read Nov 17, 2026, time per Intelligence Index task at low effort is 1.2 minutes for Terra and 5.4 minutes for Fable. At max it is 14.0 minutes against 40.8. Output speed on that same page is higher for Fable, about 105 to 112 tokens per second, against about 64 to 71 for Terra. OpenAI calls Terra a fast model and says it completes tasks over three times faster than GPT-6 Astra and GPT-6.1 Sol. That sentence does not mention Fable.

What is the context window for each model?

Anthropic's Fable 5.2 model overview lists a 1 million-token context and 128K max output. OpenAI's GPT-6.1 Terra model page lists a 1,050,000 context window, 922,000 maximum input tokens, and 128,000 max output tokens. The launch post itself does not print the window. Artificial Analysis prints 1M for both.

Can I use Claude Fable 5.2 or GPT-6.1 Terra from China?

Anthropic's availability table lists the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry. Those pages do not name a mainland China endpoint. OpenAI's Terra post lists ChatGPT Work, Codex, and the API id gpt-6.1-terra, plus Amazon Bedrock as openai.gpt-6.1-terra. This site's Images 2.6 comparison, using OpenAI's supported-countries article read Nov 17, 2026, did not find mainland China on that list. We did not open either API from inside China.

Where do the benchmark numbers come from?

Anthropic's Nov 9, 2026 table is quoted on the Claude Fable 5.2 page and is not rerun here. OpenAI's 50.3 percent Terminal-Bench 4.0 score and the error-rate drop from 53.6 percent to 16.7 percent are from OpenAI's launch post. The Intelligence Index pair, 65 and 51, is Artificial Analysis, index version 5.0.1. This page is based on official docs, pricing pages, and public reports.