> Index of everything AY Listicle publishes for agents: https://www.aylistcle.com/llms.txt

# Best AI models for coding in November 2026: 9 ranked

> The best AI models for coding in November 2026, from Claude Fable 5.2 and Opus 5.5 to GPT-6 Astra and Gemini 4 Argon, with official prices and benchmarks.

By AY Listicle team, published 2026-11-09, 21 min read. Read it on the web: https://www.aylistcle.com/blog/best-ai-coding-models-november-2026

The best AI model for coding in November 2026 depends on how big and how vague your task is. Anthropic released Claude Fable 5.2 on November 9, 2026, and on the same day put it ahead of Claude Opus 5.5, GPT-6 Astra and Gemini 4 Carbon in its own coding benchmark table.

Vendor tables are not neutral, and each vendor reports different benchmarks. So this list ranks nine models on what you can check: what each vendor says and measures about coding, the price per million tokens, the context window, and who can use it today.

Every price and spec below was read on the vendor's own page on November 9, 2026. We update this list weekly.

**TL;DR: For most coding work, start with Claude Opus 5.5, which Anthropic itself calls its "strongest daily driver". Move to Claude Fable 5.2 for large, loosely defined work across several repositories. If you live in Codex, GPT-6 Astra is the OpenAI pick, and GPT-6.1 Sol gets close to it at one fifth of the price.**

## Best AI models for coding: a quick overview

If you want the strongest model you can buy today:

- **Claude Fable 5.2**: best for large, loosely defined engineering work across several repositories, at twice Opus 5.5's token price.
- **Claude Opus 5.5**: best daily driver for agentic coding, and the model Anthropic tells most people to start with.
- **GPT-6 Astra**: best for teams that work in OpenAI's Codex and want the hardest end-to-end tasks done there.
- **Gemini 4 Argon**: best Google model you can use today, while the newer Gemini 4 Carbon is still limited to cyber defenders.

If you run coding agents all day and watch the bill:

- **GPT-6.1 Sol**: best value in Codex, which OpenAI says "nearly matches" Astra on agentic coding at one fifth of the price.
- **Claude Sonnet 5.5**: best mid-price Claude model for everyday coding.
- **Kimi K3.1 Pro**: best open-weight model, for teams that need to run the weights themselves.

If you need a cheap model for subagents and bulk work:

- **GPT-6 Luna and Claude Haiku 5.5**: best for focused, high-volume coding tasks such as subagents, at $0.10 per million input tokens.

If you do security research:

- **Claude Mythos 5.2**: best for verified cyberdefenders, and the only model here most teams cannot get.

Prices are per million tokens on the standard tier, input then output, as of November 9, 2026, from each vendor's pricing page.

| Model | Best for | Price per 1M tokens (in / out) | Context window | Where you can use it |
|---|---|---|---|---|
| Claude Fable 5.2 | Large, loosely defined engineering work | $8 / $40 | 1M | Claude Pro and up, Claude Code, Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
| Claude Opus 5.5 | Daily agentic coding | $4 / $20 | 1M | Claude Pro and up, Claude Code, Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
| GPT-6 Astra | Hardest tasks in Codex | $10 / $50 | 1.05M | ChatGPT Plus and up, Codex, OpenAI API, Azure, AWS Bedrock |
| Gemini 4 Argon | Google Cloud and Android teams | $2 / $10 until Nov 30, then $4 / $20 | 1M | Gemini app, Gemini API, AI Studio, Antigravity, Android Studio |
| GPT-6.1 Sol | Long coding runs on a budget | $2 / $10 | 1.05M | ChatGPT Work and Codex (Plus and up), OpenAI API |
| Claude Sonnet 5.5 | Everyday coding at a mid price | $2 / $10 | 1M | Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
| Kimi K3.1 Pro | Open weights you can self-host | $2 / $10 | 1M | Kimi Code, Moonshot API, Hugging Face weights |
| GPT-6 Luna and Claude Haiku 5.5 | Subagents and bulk tasks | $0.10 / $0.50 (both) | 1.05M and 1M | Codex and OpenAI API; Claude API |
| Claude Mythos 5.2 | Verified security and life sciences teams | $8 / $40 | 1M | Verified organizations only |

## How we ranked these coding models

We did not run our own benchmark for this page. We ranked on four things you can check, in this order.

First, what the vendor measured on coding, read from its own page. Anthropic, OpenAI and Google report different benchmarks, so a number from one table cannot be compared with a number from another. Where a single table puts several vendors side by side, we say whose table it is.

Second, price per million tokens. Third, the context window and maximum output. Fourth, who can use the model today.

Two independent leaderboards have now scored Fable 5.2. On the [Artificial Analysis Intelligence Index](https://artificialanalysis.ai/models/claude-fable-5-2) v5.0.1, Claude Fable 5.2 (max) scores 65 and ranks #1 of 74 (checked November 9, 2026).

Next come Gemini 4 Carbon (high) at 60, Opus 5.5 (max) at 59 and GPT-6 Astra (max) at 57. The [Vals Index](https://www.vals.ai/models/anthropic_claude-fable-5-2), a different index, put Fable 5.2 first of 36 models at 71.31% on launch day. Mythos 5.2 was not on Artificial Analysis at our check.

## 1. Claude Fable 5.2, best for large, loosely defined engineering work

Claude Fable 5.2 is Anthropic's most capable generally available model, released on November 9, 2026, for long, open-ended technical work. Anthropic's [Fable page](https://www.anthropic.com/claude/fable) calls it "our most capable model for ambitious software engineering" and says it "takes on large, loosely defined problems across multiple repositories and large monorepos".

In Anthropic's [launch table](https://www.anthropic.com/claude-fable-and-mythos-5-2), Fable 5.2 scores 69.9% on FrontierSWE v2, ahead of GPT-6 Astra (65.5%), Opus 5.5 (62.3%) and Gemini 4 Carbon (61.2%). On FrontierCode 2.0 it scores 49.1% at its highest effort setting. It does not win every row: Gemini 4 Carbon leads on AutomationBench, a business workflow test, 53.9% to 51.4%.

**Key features**
- 1M token context window and 128K max output, per the [models overview](https://platform.claude.com/docs/en/about-claude/models/overview)
- Model ID `claude-fable-5-2`, the same on the Claude API, Google Cloud, Microsoft Foundry and Claude Platform on AWS
- Fast mode in Claude Code, which Anthropic says generates up to 2.5x faster, at a higher price
- Anthropic says it "tests its own work as it goes"
- Up to 30% faster output than Fable 5.1, by [Anthropic's own claim](https://www.anthropic.com/claude-fable-and-mythos-5-2)

**Best for**
- Staff engineers handing off multi-repo changes with a loose spec
- Teams whose evals on Opus 5.5 at high effort still fall short, which is the case [Anthropic's docs](https://platform.claude.com/docs/en/about-claude/models/overview) name for moving up

**Pricing**
$8 per million input tokens and $40 per million output tokens, 20% below Fable 5.1's $10 and $50 ([CNBC](https://www.cnbc.com/2026/11/09/anthropic-claude-fable-5-2-mythos-5-2.html)). The Batch API halves that to $4 and $20. Fast mode costs $16 and $80. In the Claude apps it comes with Pro, Max, Ultra, Team and Enterprise plans, and the [launch offer](https://support.claude.com/en/articles/17506494-claude-fable-5-2-launch-offers) gives Pro and standard Team seats Fable for one week, up to 50% of weekly usage limits. (as of Nov 9, 2026, [Claude pricing](https://platform.claude.com/docs/en/about-claude/pricing))

**Pros**
- Highest score on both agentic coding benchmarks in Anthropic's table
- At high effort, Anthropic says it matches Opus 5.5's best FrontierCode score (45.3% vs 45.5%) for about half the cost per task
- 1M context fits large monorepos in one session

**Cons**
- Twice Opus 5.5's price per token, so the bill only drops if it finishes tasks in fewer steps
- 30-day data retention for safety monitoring by default, and safeguards that block high-risk cyber work
- On [Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-fable-5-2.html) only through US Geo or Global cross-Region profiles, after you opt in to AWS's `aws_review` data retention mode

## 2. Claude Opus 5.5, best daily driver for agentic coding

Claude Opus 5.5 is Anthropic's main model for everyday agentic coding, released on September 22, 2026. Anthropic's [Opus page](https://www.anthropic.com/claude/opus) calls it "our strongest Opus model for agentic coding" and lists building features, debugging, refactoring and code review in large codebases.

Anthropic's own advice on launch day was to keep using it: "For most coding and knowledge tasks, we recommend that users start with Claude Opus 5.5, our strongest daily driver" ([announcement](https://www.anthropic.com/claude-fable-and-mythos-5-2)). In the [Opus 5.5 launch post](https://www.anthropic.com/claude-opus-5-5), it scores 66.4% on Terminal-Bench 4.0 at xhigh effort.

**Key features**
- 1M token context window and 128K max output
- Model ID `claude-opus-5-5`, also on Amazon Bedrock as `anthropic.claude-opus-5-5`
- Fast mode in Claude Code and on the Claude Platform
- Available with zero data retention

**Best for**
- Developers who use Claude Code every day on real codebases
- Bedrock teams whose requests must stay in the EU, Australia or Japan
- Companies that need zero data retention

**Pricing**
$4 per million input tokens and $20 per million output tokens. Batch is $2 and $10, fast mode is $8 and $40. In the apps it comes with Pro, Max, Ultra, Team and Enterprise. (as of Nov 9, 2026, [Claude pricing](https://platform.claude.com/docs/en/about-claude/pricing))

**Pros**
- Half Fable 5.2's token price, with a FrontierSWE v2 score of 62.3% in Anthropic's table
- Runs on Amazon Bedrock, Google Cloud, Microsoft Foundry and the Claude API ([Opus docs](https://platform.claude.com/docs/en/models/opus-5-5/overview))
- GitHub's CPO said that across Copilot CLI and VS Code it "used among the fewest tokens and steps we measured" ([Anthropic](https://www.anthropic.com/claude-opus-5-5))

**Cons**
- Trails Fable 5.2 by 7.6 points on FrontierSWE v2 and 3.6 points on FrontierCode 2.0 in Anthropic's table
- It is "no longer available with 'thinking' mode switched off" ([Anthropic](https://www.anthropic.com/claude-opus-5-5))

## Opus 5.5 vs Fable 5.2 for coding: which should you use?

Use Opus 5.5 by default and switch to Fable 5.2 when a task is big, vague and spans several repositories. That is Anthropic's own advice, and the numbers in its [launch table](https://www.anthropic.com/claude-fable-and-mythos-5-2) back it up.

| | Claude Opus 5.5 | Claude Fable 5.2 |
|---|---|---|
| Price per 1M tokens (in / out) | $4 / $20 | $8 / $40 |
| FrontierSWE v2 | 62.3% | 69.9% |
| FrontierCode 2.0, best score | 45.5% | 49.1% (xhigh effort) |
| FrontierCode 2.0, Fable at high effort | 45.5% | 45.3%, about half the cost per task |
| Context and max output | 1M / 128K | 1M / 128K |
| Amazon Bedrock | `anthropic.claude-opus-5-5` | `anthropic.claude-fable-5-2`, cross-Region profiles only |
| Data retention | Zero retention available | 30 days by default |

The surprising line is the fourth one. Fable 5.2 costs twice as much per token, yet on FrontierCode at high effort Anthropic says it reaches Opus 5.5's best score for about half the cost per task. It gets there in fewer steps. That is one benchmark from the vendor, so test it on your own repository before you switch a team.

## 3. GPT-6 Astra, best for the hardest tasks in Codex

GPT-6 Astra is OpenAI's flagship model for complex reasoning and coding, released on September 3, 2026. OpenAI's [launch post](https://openai.com/index/gpt-6-astra/) says "GPT‑6 Astra is the best model for software engineering to date", and its [models page](https://developers.openai.com/api/docs/models) calls it "our flagship model for complex reasoning and coding".

In [OpenAI's own table](https://openai.com/index/gpt-6-astra/), Astra scores 57.9% on Terminal-Bench 4.0 and 74.1% on DeepSWE v1.1, but that table compares it with Claude Fable 5.1 and Opus 5, not the current Claude models. In Anthropic's newer table it scores 65.5% on FrontierSWE v2, behind Fable 5.2.

**Key features**
- 1,050,000 token context window and 128K max output ([model page](https://developers.openai.com/api/docs/models/gpt-6-astra))
- Runs in the Codex desktop app, web, CLI (version 0.153.0 or newer) and IDE extension
- Also on Microsoft Azure and AWS Bedrock as `openai.gpt-6-astra`
- Knowledge cutoff April 30, 2026

**Best for**
- Teams already working in ChatGPT and Codex
- End-to-end tasks that mix code, browsing and computer use, which OpenAI names as Astra's strengths

**Pricing**
$10 per million input tokens and $50 per million output tokens, with cached input at $1. Prompts over 272K tokens cost 2x on input and 1.5x on output. Plus and Business Standard seats include limited Astra usage in Codex. (as of Nov 9, 2026, [OpenAI pricing](https://developers.openai.com/api/docs/pricing))

**Pros**
- OpenAI's strongest coding model, available on Azure and Bedrock as well as OpenAI's own API
- Deep Codex integration across app, CLI and IDE

**Cons**
- The most expensive model per token on this list
- OpenAI says it "will refuse to comply with more advanced cybersecurity tasks", and safety checks "can sometimes interrupt legitimate work"
- OpenAI's [GPT-6 guide](https://developers.openai.com/api/docs/guides/latest-model) warns it can write "broader tests than the task requires"

## 4. Gemini 4 Argon, best Google model you can use today

Gemini 4 Argon is Google's current generally available Gemini model for coding and agents, announced on September 30, 2026 and opened to developers on October 13. Google's [pricing page](https://aistudio.google.com/docs/pricing#gemini-4-argon) says it is "built for complex software engineering, deep mathematical and scientific problem solving, autonomous multi-step agents".

Google's newer flagship, Gemini 4 Carbon, was [announced on October 28](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/), but only "a subset of cyber defenders" can use it so far, and Google gives no API model ID yet. In Anthropic's table, Carbon scores 61.2% on FrontierSWE v2. Until it opens up, Argon is the Gemini you can build with.

**Key features**
- Model ID `gemini-4-argon`, 1,048,576 token context shared between input and output ([model page](https://aistudio.google.com/docs/models/gemini-4-argon))
- 65,536 max output by default, more with long decode
- 77.9% on DeepSWE v1.1 in Google's [model card](https://deepmind.google/models/model-cards/gemini-4-argon/), ahead of Opus 5.5 (74.2%) in that table
- In the Gemini app, Gemini API, Google AI Studio, Google Antigravity and Android Studio

**Best for**
- Teams already building on Google Cloud and AI Studio
- Android developers who want the model inside Android Studio

**Pricing**
$2 per million input tokens and $10 per million output tokens through November 30, 2026, rising to $4 and $20 on December 1. Not on the Gemini API free tier. (as of Nov 9, 2026, [Gemini API pricing](https://aistudio.google.com/docs/pricing#gemini-4-argon))

**Pros**
- Leads Google's own table on DeepSWE v1.1
- Introductory price matches Sonnet 5.5 and GPT-6.1 Sol until the end of November

**Cons**
- Google's own model card puts it behind Opus 5.5 on FrontierSWE v2 (55.0% vs 62.3%) and Terminal-Bench 4.0 (57.4% vs 66.4%)
- The price doubles on December 1, 2026
- Knowledge cutoff March 2026, and the model card warns of "occasional slowness or timeout issues"

## 5. GPT-6.1 Sol, best value for long coding runs in Codex

GPT-6.1 Sol is OpenAI's mid-priced model for repeated, long-running work across code, apps and documents, released on September 29, 2026. OpenAI's [launch post](https://openai.com/index/introducing-gpt-6-1-sol/) says it "nearly matches GPT-6 Astra's intelligence on agentic coding, computer use, and professional work at one-fifth of Astra's standard input and output token prices".

It replaced GPT-5.5 as the recommended Codex model on paid plans. OpenAI no longer ships a separate Codex-named model: `gpt-5.3-codex` is deprecated, and the [deprecations page](https://developers.openai.com/api/docs/deprecations) names GPT-6 Sol as its replacement.

**Key features**
- 1,050,000 token context window and 128K max output; Codex defaults to a 500K window ([model page](https://developers.openai.com/api/docs/models/gpt-6.1-sol))
- In Codex app, CLI and IDE extension, and in ChatGPT Work
- On DeepSWE v1.1, OpenAI says it matches Astra "at roughly one-fifth of the cost"
- Ultrafast mode at 6x the standard price

**Best for**
- Teams running Codex agents for hours a day who need to watch spend
- Developers who want most of Astra's coding ability at Sonnet-level token prices

**Pricing**
$2 per million input tokens and $10 per million output tokens, with cached input at $0.10. Included for Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex. (as of Nov 9, 2026, [OpenAI pricing](https://developers.openai.com/api/docs/pricing))

**Pros**
- One fifth of Astra's token price for close to the same agentic coding results, by OpenAI's measure
- Same 1.05M context window as Astra

**Cons**
- Not available in regular ChatGPT chat yet, only in ChatGPT Work and Codex
- OpenAI publishes its 6.1 Sol scores in charts only, with no standalone number in the text

## 6. Claude Sonnet 5.5, best mid-price Claude model for everyday coding

Claude Sonnet 5.5 is Anthropic's middle model, released on September 28, 2026, and described in the [models overview](https://platform.claude.com/docs/en/about-claude/models/overview) as "the best combination of speed and intelligence". Anthropic's [Sonnet 5.5 post](https://www.anthropic.com/claude-sonnet-5-5) reports 70.6% on Terminal-Bench 4.0, up from 10.3% for Sonnet 5.

The same post is open about its limit: "Opus 5.5 remains clearly stronger at complex, open-ended work requiring sustained judgment." Read that as a clear split. Sonnet for well-scoped tickets, Opus or Fable for the hard ones.

**Key features**
- 1M token context window and 128K max output
- Model ID `claude-sonnet-5-5`
- On the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry
- Half Opus 5.5's token price

**Best for**
- Teams that run many parallel coding agents on well-scoped tasks
- Products that call a model inside an editor or CI step, where speed matters

**Pricing**
$2 per million input tokens and $10 per million output tokens. (as of Nov 9, 2026, [Claude pricing](https://platform.claude.com/docs/en/about-claude/pricing))

**Pros**
- Strong Terminal-Bench 4.0 result for its price, by Anthropic's measure
- Same 1M context as Opus and Fable

**Cons**
- Anthropic itself says Opus 5.5 is clearly stronger on open-ended work
- Not in Anthropic's five-way coding table, so there is no direct comparison with Fable 5.2 or GPT-6 Astra

## 7. Kimi K3.1 Pro, best open-weight model for coding

Kimi K3.1 Pro is Moonshot AI's open-weight model, released on October 23, 2026, which its [launch post](https://www.kimi.ai/blog/kimi-k3-1-pro) calls "our most capable 3T-class open-weight model to date". It is the only model on this list whose weights you can download and run yourself.

Moonshot's [quickstart](https://platform.kimi.ai/docs/guide/kimi-k3-1-pro-quickstart) says it scores 65.7% on Terminal-Bench 4.0, "ahead of GPT-6 Astra (58.2%) and narrowly behind Claude Opus 5.5 (66.4%)". Those are Moonshot's numbers, and Moonshot notes that benchmarks against proprietary models "may involve some fallback behavior".

**Key features**
- 1,048,576 token context window
- Weights on Hugging Face as moonshotai/Kimi-K3.1-Pro, 2.8T total parameters with 104B active
- Works in Kimi Code and through the Moonshot API
- A highspeed variant at a higher price

**Best for**
- Teams that must self-host for compliance or cost reasons
- Researchers who want to fine-tune a strong coding model

**Pricing**
$2 per million input tokens and $10 per million output tokens on the Moonshot API, with cached input at $0.20. The highspeed variant costs $4 and $20, and batch jobs cost 60% of the standard price. (as of Nov 9, 2026, [Moonshot pricing](https://platform.moonshot.ai/docs/pricing/chat))

**Pros**
- Open weights under the Kimi K3.1 Pro License, close to Opus 5.5 on Terminal-Bench 4.0 by Moonshot's measure
- Same API price as Sonnet 5.5 and GPT-6.1 Sol

**Cons**
- The license adds conditions for large "Model as a Service" businesses
- At 2.8T parameters, self-hosting needs serious GPU capacity
- Moonshot says its "cyber capability remains behind that of frontier proprietary models"

## 8. GPT-6 Luna and Claude Haiku 5.5, best for subagents and bulk coding tasks

GPT-6 Luna and Claude Haiku 5.5 are the two cheapest coding-capable models from OpenAI and Anthropic, both starting at $0.10 per million input tokens. We list them together because they do the same job: run many small, well-defined tasks under a bigger model.

OpenAI's [Codex models page](https://developers.openai.com/codex/models) calls Luna "our most efficient model for focused, high-volume tasks, including summarization, extraction, and focused coding". Anthropic's [Haiku 5.5 post](https://www.anthropic.com/claude-haiku-5-5) says Haiku "pairs well with Opus 5.5 and Sonnet 5.5 as a subagent on coding work".

**Key features**
- GPT-6 Luna: 1,050,000 token context, 128K max output, released September 22, 2026 ([model page](https://developers.openai.com/api/docs/models/gpt-6-luna))
- OpenAI says Luna at max effort scores 66.6% on DeepSWE v1.1, "comparable to Claude Opus 5 and Fable 5 at medium effort" ([OpenAI](https://openai.com/index/introducing-gpt-6-sol-and-luna/))
- Claude Haiku 5.5: 1M token context, 128K max output, released October 7, 2026
- Both vendors position them as helpers to a larger coding model

**Best for**
- Subagents that search, summarize or edit one file at a time
- Bulk jobs such as test generation or migrations across many small files

**Pricing**
GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens. Claude Haiku 5.5 starts at the same $0.10 and $0.50, rising to $0.50 and $2.50 for prompts over 100K tokens. (as of Nov 9, 2026, [OpenAI pricing](https://developers.openai.com/api/docs/pricing), [Claude pricing](https://platform.claude.com/docs/en/about-claude/pricing))

**Pros**
- About 1/40 of Opus 5.5's input price
- Free and Go ChatGPT users can use Luna in the Codex desktop app, per OpenAI

**Cons**
- Built for tasks where you know what a good result looks like, not for open-ended design work
- Luna does not support the Ultra reasoning level, and Enterprise admins must switch it on first

## 9. Claude Mythos 5.2, best for verified security and life sciences teams

Claude Mythos 5.2 is the same underlying model as Fable 5.2, without the cyber and biology safeguards Anthropic adds to Fable, released on November 9, 2026 for vetted organizations only. Anthropic's [Mythos page](https://www.anthropic.com/claude/mythos) calls it "our most capable model for cybersecurity and life sciences research".

Most readers cannot pick it. The [Mythos docs](https://platform.claude.com/docs/en/models/mythos-5-2/overview) say it "is available only to organizations verified through Anthropic's verification programs, such as the Cyber Verification Program". Claude Enterprise customers can use Claude Security, which runs on the model family that includes Mythos 5.2.

**Key features**
- 1M token context window and 128K max output, the same specifications as Fable 5.2
- 89.6% on CVE Rediscovery in Anthropic's [launch table](https://www.anthropic.com/claude-fable-and-mythos-5-2), up from 78.3% for Mythos 5.1
- 75.3% on Terminal-Bench-Science 0.1
- On the Claude API, Google Cloud and Microsoft Foundry for verified organizations

**Best for**
- Security teams doing vulnerability research
- Life sciences labs in Anthropic's Life Sciences Verification Program

**Pricing**
$8 per million input tokens and $40 per million output tokens, the same as Fable 5.2, with fast mode only for verified organizations. (as of Nov 9, 2026, [Claude pricing](https://platform.claude.com/docs/en/about-claude/pricing))

**Pros**
- The only model here that Anthropic offers to security teams without Fable's cyber safeguards
- Same coding ability as Fable 5.2, because it is the same model

**Cons**
- Not available to the public: you need to pass a verification program
- 30-day data retention for safety monitoring by default

## How to choose the best AI model for coding

The right pick depends on four questions. Answer them in order.

### 1) Is the task well scoped or open-ended?

If you can write the ticket in three lines and say what done looks like: pick Claude Sonnet 5.5 or GPT-6.1 Sol. Both cost $2 and $10 per million tokens. If the task is "split this monolith into services" across several repositories: pick Claude Fable 5.2 or GPT-6 Astra, because both vendors position these for open-ended work.

### 2) Which coding agent does your team already use?

If your team works in Claude Code: start with Opus 5.5 and keep Fable 5.2 for the hard tickets. If it works in Codex: start with GPT-6.1 Sol and move up to GPT-6 Astra. If it builds on Google Cloud or Android: Gemini 4 Argon is the Gemini you can use now. Switching agents costs more time than any model upgrade saves.

### 3) Are you paying per token or per seat?

On a Claude or ChatGPT subscription, the price per token matters less than the usage limits. Check the [Fable 5.2 launch offer](https://support.claude.com/en/articles/17506494-claude-fable-5-2-launch-offers) before you upgrade a plan. On the API, count cost per finished task, not cost per token: Anthropic's own FrontierCode result shows a model at twice the token price finishing for about half the cost per task.

### 4) Where must your data stay?

If you need Amazon Bedrock, Fable 5.2 is there as `anthropic.claude-fable-5-2`, but AWS offers it only through the `us.` and `global.` cross-Region profiles ([Bedrock model card](https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-fable-5-2.html)). If requests must stay in the EU, Australia or Japan, use Opus 5.5: it has `eu.`, `au.` and `jp.` profiles and Fable 5.2 does not yet ([Opus 5.5 Bedrock card](https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-opus-5-5.html)). If you need zero data retention, Opus 5.5 offers it, while Fable 5.2 keeps data 30 days for safety monitoring by default.

Run a two-week trial on your own repository with two models from this list. Measure merged pull requests and cost per merged change, and ignore the leaderboard until you have your own numbers.

Picking a model is half the job. If you build a developer tool, the other half is being named when buyers ask ChatGPT or Claude which one to use, and those answers often lean on "best of" lists like this one. [Request a spot on AY Listicle](https://www.aylistcle.com/signup) to get your product into lists that buyers and AI search read, or read [how getting listed with credits works](/blog/getting-listed-with-credits).

## FAQ

**What is the best AI model for coding in November 2026?**
Claude Fable 5.2 has the highest agentic coding scores in Anthropic's launch table, with 69.9% on FrontierSWE v2. For most daily coding, Anthropic itself recommends starting with Claude Opus 5.5, which costs half as much per token. OpenAI users should look at GPT-6 Astra and GPT-6.1 Sol in Codex.

**What is the best Claude model for coding?**
Claude Opus 5.5 is the best Claude model for most coding, and Claude Fable 5.2 is the best for large, loosely defined work. Anthropic recommends Opus 5.5 as the starting point and Fable 5.2 when Opus 5.5 at high effort still falls short. Sonnet 5.5 is the cheaper option for well-scoped tasks.

**Is Claude Fable 5.2 better than Opus 5.5 for coding?**
Claude Fable 5.2 scores higher than Opus 5.5 on both coding benchmarks in Anthropic's table: 69.9% vs 62.3% on FrontierSWE v2 and 49.1% vs 45.5% on FrontierCode 2.0. It also costs twice as much per token. Anthropic says that at high effort Fable 5.2 matches Opus 5.5's best FrontierCode score for about half the cost per task.

**How much does Claude Fable 5.2 cost?**
Claude Fable 5.2 costs $8 per million input tokens and $40 per million output tokens on the Claude API, as of November 9, 2026. That is 20% less than Fable 5.1's $10 and $50. Pages that still list $10 and $50 are showing the old model's price.

**Can I use Claude Mythos 5.2?**
Claude Mythos 5.2 is available only to organizations that pass an Anthropic verification program, such as the Cyber Verification Program. It is the same model as Fable 5.2 without Fable's cyber and biology safeguards. If you are not a vetted security or life sciences team, Fable 5.2 is the version you can use.

**Is Claude Fable 5.2 on Artificial Analysis yet?**
Yes. Claude Fable 5.2 (max) ranks #1 of 74 on the Artificial Analysis Intelligence Index v5.0.1, with a score of 65 ([Artificial Analysis](https://artificialanalysis.ai/models/claude-fable-5-2), checked November 9, 2026). It costs $26.54 per index task, against $10.04 for Opus 5.5 (max) at 59. The Vals Index, a different leaderboard, ranked Fable 5.2 first of 36 models at 71.31% on launch day.

**Claude or GPT for coding?**
Claude Fable 5.2 leads GPT-6 Astra on FrontierSWE v2 in Anthropic's table, 69.9% to 65.5%, but that is Anthropic's test. OpenAI's own table, run against older Claude models, shows Astra ahead on Terminal-Bench 4.0. Pick by the agent your team uses: Claude Code for Claude, Codex for GPT.

**Which AI coding model is cheapest?**
GPT-6 Luna and Claude Haiku 5.5 both start at $0.10 per million input tokens and $0.50 per million output tokens. Both vendors position them for focused, high-volume tasks such as subagents. For open-ended work, the cheapest strong options are GPT-6.1 Sol and Claude Sonnet 5.5 at $2 and $10.

*Sources: [Anthropic, Introducing Claude Fable 5.2 and Claude Mythos 5.2](https://www.anthropic.com/claude-fable-and-mythos-5-2), [Claude models overview](https://platform.claude.com/docs/en/about-claude/models/overview), [Claude pricing](https://platform.claude.com/docs/en/about-claude/pricing), [Claude Fable 5.2 launch offers](https://support.claude.com/en/articles/17506494-claude-fable-5-2-launch-offers), [OpenAI, GPT-6 Astra](https://openai.com/index/gpt-6-astra/), [OpenAI models](https://developers.openai.com/api/docs/models), [OpenAI pricing](https://developers.openai.com/api/docs/pricing), [Codex models](https://developers.openai.com/codex/models), [Gemini API pricing](https://aistudio.google.com/docs/pricing), [Gemini 4 Argon model card](https://deepmind.google/models/model-cards/gemini-4-argon/), [Introducing Gemini 4 Carbon](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/), [Kimi K3.1 Pro](https://www.kimi.ai/blog/kimi-k3-1-pro), [Artificial Analysis leaderboard](https://artificialanalysis.ai/leaderboards/models), [Artificial Analysis, Claude Fable 5.2](https://artificialanalysis.ai/models/claude-fable-5-2), [Amazon Bedrock, Claude Fable 5.2 model card](https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-fable-5-2.html), [Vals AI](https://www.vals.ai/models/anthropic_claude-fable-5-2). All checked November 9, 2026.*