πŸ† TopRankLand
← All Rankings
Software

Best AI Coding Assistants 2026

Ranking the top AI coding tools and assistants that help developers write, refactor, and ship code faster in 2026.

Last updated: 2026-10-06 Β· 13 entries tracked daily

Rank Trend β€” Top 10

Lower = better rank. Showing last 133 days.

Current Rankings

#1
Claude Code Anthropic
$17/mo 9.6/10

Terminal-first AI coding agent now powered by Claude Opus 4.7, scoring 87.6% on SWE-bench Verified β€” the highest of any commercial tool β€” with a 1M-token context window and autonomous multi-file refactoring across massive codebases.

Code Quality & Accuracy 9.8
Context & Codebase Understanding 9.7
IDE Integration & UX 8.5
Agentic / Autonomous Capability 9.8
Value for Money 9.2
#2
$20/mo+ 9.4/10

Cloud-based autonomous coding agent from OpenAI, bundled with ChatGPT Plus and Pro plans, capable of parallel task execution in isolated environments.

Code Quality & Accuracy 9.2
Context & Codebase Understanding 8.2
IDE Integration & UX 8.2
Agentic / Autonomous Capability 9.6
Value for Money 8.0
#3
Cursor Anysphere
$20/mo 9.1/10

A VS Code fork rebuilt around AI; Composer 2 added parallel agent execution to its already strong multi-file editing, and the deep codebase indexing still leads the IDE pack for project-level understanding.

Code Quality & Accuracy 9.0
Context & Codebase Understanding 8.5
IDE Integration & UX 9.6
Agentic / Autonomous Capability 9.3
Value for Money 8.1
#4
Free / usage-based 8.5/10

Google's agent-first IDE launched at I/O 2026, pairing a multi-agent desktop app with a Go CLI and SDK, all driven by Gemini 3.5 Flash. Flash runs roughly four times faster than frontier rivals while holding a huge context window, and the free public tier makes it the easiest way I've found to run parallel coding agents without a subscription.

Code Quality & Accuracy 8.2
Context & Codebase Understanding 8.8
IDE Integration & UX 8.5
Agentic / Autonomous Capability 9.0
Value for Money 8.8
#5
Bundled (AI Pro/Ultra) 8.2/10

Google's async autonomous coding agent powered by Gemini 2.5 Pro; integrates natively with GitHub to plan and execute multi-file tasks in a secure cloud sandbox with no IDE required.

Code Quality & Accuracy 8.0
Context & Codebase Understanding 8.5
IDE Integration & UX 7.5
Agentic / Autonomous Capability 8.5
Value for Money 8.5
#6
Devin Cognition
$20/mo+ 8.0/10

The world's first fully autonomous AI software engineer, capable of independently planning, executing, and deploying complete coding tasks with minimal human input.

Code Quality & Accuracy 8.0
Context & Codebase Understanding 8.0
IDE Integration & UX 7.5
Agentic / Autonomous Capability 9.0
Value for Money 7.0
#7
GitHub Copilot GitHub / Microsoft
$10/mo 8.0/10

The industry-standard AI code completion tool used by 15+ million developers, with broad IDE support and a genuinely useful free tier.

Code Quality & Accuracy 8.5
Context & Codebase Understanding 8.0
IDE Integration & UX 9.5
Agentic / Autonomous Capability 8.2
Value for Money 6.8
#8
Kiro AWS
Free / $19/mo 7.9/10

AWS's spec-driven agentic IDE, built on a VS Code base and powered by Bedrock, now the home for what used to be the Amazon Q Developer CLI. Its spec-first workflow and Agent Hooks automation turn requirements into structured plans before code gets written, which is why I reach for it on enterprise projects where traceability matters as much as speed.

Code Quality & Accuracy 8.0
Context & Codebase Understanding 8.0
IDE Integration & UX 8.5
Agentic / Autonomous Capability 8.3
Value for Money 8.0
#9
Amazon Q Developer Amazon Web Services
$19/mo 7.8/10

AWS's AI coding assistant with deep cloud and infrastructure integration, supporting code generation, security scanning, and automated framework migrations.

Code Quality & Accuracy 7.5
Context & Codebase Understanding 8.0
IDE Integration & UX 8.5
Agentic / Autonomous Capability 7.5
Value for Money 8.0
#10
Zed Zed Industries
Free+ 7.8/10

A high-performance, GPU-accelerated AI-native code editor built in Rust, offering fast AI chat and inline assistance using your own API key.

Code Quality & Accuracy 8.0
Context & Codebase Understanding 7.5
IDE Integration & UX 8.5
Agentic / Autonomous Capability 7.0
Value for Money 8.5
#11
Aider Paul Gauthier (OSS)
Free (API cost) 7.7/10

A free open-source terminal-based AI pair programmer that works with 100+ LLMs, auto-commits changes with Git, and supports multi-file editing via the command line.

Code Quality & Accuracy 8.0
Context & Codebase Understanding 8.0
IDE Integration & UX 7.0
Agentic / Autonomous Capability 8.0
Value for Money 9.5
#12
Windsurf Codeium (acq. Cognition AI)
$15/mo 7.7/10

Agentic IDE with 1M+ active users; acquired by Cognition AI (makers of Devin) in Dec 2025, combining two of the top autonomous coding tools under one parent. Known for clean UX and competitive $15/mo pricing.

Code Quality & Accuracy 7.5
Context & Codebase Understanding 8.0
IDE Integration & UX 8.5
Agentic / Autonomous Capability 7.5
Value for Money 8.0
#13
$20/mo+ 7.6/10

Enterprise-focused AI coding assistant with deep codebase indexing, supporting VS Code and JetBrains with credit-based usage across autonomous agent tasks.

Code Quality & Accuracy 7.5
Context & Codebase Understanding 8.0
IDE Integration & UX 8.0
Agentic / Autonomous Capability 7.5
Value for Money 7.5

Today's Analysis Β· 2026-10-06

No movement in the rankings today, and the broader data I found this week actually reinforces why the top three sit where they do. Independent coverage now pegs AI coding assistant adoption at 85% of professional developers with teams reporting roughly 40% productivity gains, and the dominant usage pattern described across multiple outlets is a hybrid one, Cursor or Copilot for fast inline editing during the day, Claude Code for the harder multi-file, multi-step work. That hybrid pattern is exactly why Claude Code keeps the top spot here, its agentic capability score of 9.8 and context window score of 9.7 are built for exactly the kind of autonomous, read-the-whole-codebase task that the other tools in this list explicitly delegate to it instead of competing on. OpenAI Codex holds second on the strength of its own 9.6 agentic score, close enough to Claude Code that the gap genuinely comes down to code quality and IDE integration more than raw capability. Cursor's 9.6 IDE integration score remains unmatched in this list, which is why it stays the top choice for anyone who wants an AI-native editor instead of a terminal-first agent, and that's a real product difference, not a ranking technicality. Nothing from Google Antigravity, Devin, or the rest of the mid-pack moved enough this week to shuffle positions, this remains a category where the fundamentals, context handling, autonomous multi-step execution, and how deeply a tool integrates into your existing workflow, decide the outcome more than any single feature announcement.

Claude Code's lead matches how developers actually use it

Multiple independent write-ups this week describe the same hybrid pattern, a fast IDE-native tool for daily edits plus Claude Code for the harder autonomous work, and that's precisely the gap its 9.8 agentic capability score and 9.7 context window score are built to fill.

OpenAI Codex earns second on genuine agentic strength, not hype

Its 9.6 agentic capability score sits within a tenth of Claude Code's, and the gap between them comes down to code quality and value for money more than any gap in autonomous task handling.

Cursor's IDE integration score of 9.6 is still unmatched here

If you want to live inside a single AI-native editor instead of bouncing to a terminal agent, Cursor's integration depth remains the best reason to pick it over tools that score higher on raw agentic capability.

The 85% adoption stat confirms this category has matured past hype

When the majority of professional developers are using these tools daily and reporting real productivity gains, rankings should move on demonstrated capability shifts, not announcements, which is exactly why nothing here reshuffled today.

References

Update History

2026-10-05

Claude Code holds the top spot today, and it's getting stronger support from fresh benchmark data: it leads SWE-bench Verified at roughly 80.8%, ahead of every editor-embedded competitor in this ranking, while Anthropic's October release of Sonnet 5.5 brings a model that's 30% faster at the same coding tasks. I'm nudging its score up slightly today, from 9.5 to 9.6, and bumping its agentic capability score from 9.7 to 9.8 to reflect that speed gain directly, since agentic workflows live and die on how many tool calls you can run per minute. OpenAI Codex holds a close second on the strength of GPT-6 Astra's coding gains, still the tightest race at the top of this entire list. Cursor holds third as the best all-around IDE experience, routing each task to whichever frontier model performs best for it, which is exactly why its ideIntegration score stays the highest in this ranking. Google Antigravity holds fourth as Google's newest entrant, still building the track record that would let it challenge the top three. GitHub Copilot remains the default pick for large enterprises buying through GitHub itself, even though its per-seat economics get expensive once you add GitHub Enterprise Cloud on top. Kiro holds eighth as AWS's spec-driven rebuild of Amazon Q Developer, a genuinely different design philosophy than anything else in this list. I'm holding every other rank steady; this week's news reinforces why the current order holds.

Claude Code's SWE-bench Lead Gets Sharper

An 80.8% SWE-bench Verified score, ahead of every editor-embedded rival here, plus a 30% faster Sonnet 5.5 release, earns today's small bump to 9.6.

Codex Keeps the Top Two Tight

GPT-6 Astra's coding gains keep OpenAI Codex within a tenth of a point of first place, the closest race in this entire ranking.

Cursor Wins by Not Picking One Model

Routing each task to whichever frontier model performs best keeps its IDE integration score the highest in this list.

Kiro's Spec-Driven Rebuild Stands Apart

AWS's ground-up replacement for Amazon Q Developer takes a genuinely different design philosophy than any other tool here.

2026-10-04

Claude Code keeps the top spot, and Anthropic gave it real reasons to stay there this week. October 1 brought mods, TypeScript based custom behavior, new UI, and feature replacements you can drop into the CLI or desktop app, plus Claude for Government, which brings FedRAMP High access with spend controls and audit logs for federal and state teams. The underlying 2.1.288 release also tightened resume and recovery flows and permission handling, the kind of reliability work that matters more once you are running agentic sessions for hours at a time. OpenAI Codex holds second on raw agentic capability, its 9.6 score there still edges Claude Code, but its context window and IDE integration scores lag behind, which is why it has not closed the overall gap. Cursor stays third as the model agnostic choice, you can point it at whichever frontier model you already pay for while keeping a familiar VS Code feel, and its 9.6 IDE integration score is still the best on this list. Google Antigravity and Jules hold fourth and fifth, both benefiting from Google folding more agentic tooling into its coding stack this fall. The broader story worth watching is consolidation, multiple outlets are now describing Cursor, Claude Code, and Codex as converging toward one shared workflow pattern instead of three distinct categories, which tracks with how similar their agentic scores have become. Nothing here justifies moving anyone's rank yet, the gaps at the top are feature depth and governance, not raw capability.

Claude Code Adds Mods and Government Access

October 1 shipped TypeScript based mods for custom UI and behavior plus Claude for Government, bringing FedRAMP High access with audit logs and spend controls for federal teams.

Codex Still Leads on Pure Agentic Capability

OpenAI Codex posts a 9.6 agentic capability score, the highest on this list, though a thinner context window and weaker IDE integration keep it behind Claude Code overall.

Cursor Remains the Best IDE Experience

A 9.6 IDE integration score, the top mark here, comes from staying model agnostic inside a familiar VS Code shell instead of locking you into one vendor's models.

The Big Three Are Starting to Converge

Industry coverage now frames Cursor, Claude Code, and Codex as merging toward one shared agentic workflow, which lines up with how close their agentic capability scores have become.

2026-10-03

Claude Code keeps the top spot, and the clearest validation this cycle is a pure cost and capability comparison. Claude Opus 5.5 now benchmarks as the strongest model available for autonomous multi file work and coding specifically, and it runs for well under half the API price of GPT-6 Astra, the model behind OpenAI Codex's second place finish. That combination shows up directly in Claude Code's 9.2 value for money score alongside a 9.8 code quality score, proof that capability and price can move in the same direction. OpenAI didn't stand still either, shipping an Agents API that lets any developer build long running coding agents that use tools, manage files, and execute code over extended sessions, the kind of infrastructure that should keep pushing Codex's 9.6 agentic capability score higher through the rest of this year. Cursor holds third with the best IDE integration score in this entire list at 9.6, and that's still the right call for anyone whose workflow depends on tight editor integration over raw model strength. Google Antigravity and Google Jules round out the top five, both built around agent orchestration, a different job than the single shot code generation Claude Code and Cursor focus on. If you're picking a coding assistant today, the model underneath still matters more than the wrapper around it, and right now that model is Opus 5.5.

Opus 5.5 Wins on Cost and Capability

Claude Opus 5.5 benchmarks as the strongest coding model available and runs for well under half the API price of GPT-6 Astra, the model behind OpenAI Codex.

OpenAI's Agents API Raises the Ceiling

New infrastructure for long running coding agents that manage files and execute code should keep pushing Codex's agentic capability score higher this year.

Cursor Still Wins on Integration

A 9.6 IDE integration score, the best in this ranking, keeps Cursor the right call for anyone whose workflow depends on tight editor tooling.

Google's Entries Play a Different Game

Antigravity and Jules are built around agent orchestration, a different job than the single shot code generation that tops this list.

2026-10-01

Claude Code keeps the top spot, and Anthropic gave me a concrete reason to reconfirm it this week. Claude Code v2.1.280 made Opus 5.5 the default model the same day Anthropic shipped it, September 22, and in autonomous multi-file work, code quality, and token efficiency, Opus 5.5 is the strongest model I've tested for agentic coding. Anthropic now ships Claude Code for the terminal, VS Code, JetBrains, the web, GitHub Actions, Slack, and mobile, and a tool that meets you everywhere you actually work beats one that's excellent in a single IDE.

OpenAI Codex holds second because GPT-6 Sol and GPT-6 Luna landed that same afternoon, with Sol becoming the default replacement for GPT-5.5 on paid Codex plans. Codex still edges out pure IDE tools on raw agentic autonomy, which keeps it ahead of Cursor here.

Cursor keeps third on IDE feel. Inline completions and multi-model flexibility remain its real strength, and plenty of developers will rightly pick it over Claude Code when they want a tight, fast loop for iterating file by file with their own hands on the keyboard.

GitHub Copilot's biggest move this cycle was structural: Copilot Pro now lets Claude Code and Codex run inside the Copilot experience alongside GitHub's own cloud agents and PR review. That's a smart bet on owning distribution for every agent developers already use, and it's why Copilot holds its rank on free tier and IDE breadth.

I'm keeping every rank the same this cycle. The model upgrades confirm the order that was already correct.

Claude Code's multi-surface reach is the real moat

Terminal, every major IDE, the web, GitHub Actions, Slack, and mobile are all covered now, and meeting developers wherever they already work matters more than a single peak benchmark score.

Opus 5.5 made the agentic gap wider

Shipped September 22 and made the Claude Code default the same day, Opus 5.5 leads on autonomous multi-file work, code quality, and token efficiency, exactly the traits that matter for agentic coding.

Codex's Sol upgrade keeps OpenAI in the autonomy race

GPT-6 Sol became the default replacement for GPT-5.5 on paid Codex plans the same day as Anthropic's release, and Codex still edges out IDE-first tools on raw agentic autonomy.

Cursor is still the right call for IDE purists

Inline completions and multi-model flexibility are Cursor's real strengths, and developers who want a fast, hands-on loop for iterating file by file should pick it over a standalone agent.

Copilot is betting on being everyone's distribution layer

Hosting Claude Code and Codex inside Copilot Pro alongside GitHub's own agents is a smart structural play, and it's exactly why Copilot holds its rank on free tier access and IDE breadth.

2026-09-30

Claude Code keeps first and OpenAI Codex keeps second today, and the benchmark update I want to flag actually shows the gap between them holding steady. Anthropic shipped Claude Opus 5.5 on September 22 at four dollars and twenty dollars per million tokens for input and output, with a one million token context window, and it became Claude Code's default Opus model the same day inside version 2.1.280. On Artificial Analysis's independent Coding Agent Index, Claude Code with Opus 5.5 now scores 66 against Codex with GPT-6 Astra at 62, a four point spread on a benchmark built to measure exactly the agentic coding work I weigh heaviest in this table. OpenAI answered with its own releases this month: GPT-6 Astra on September 3 at ten and fifty dollars per million tokens, then GPT-6 Sol at two and ten dollars and GPT-6 Luna at ten and fifty cents on September 22, with Sol positioned as the direct replacement for GPT-5.5 at half its old promotional price. That pricing move helps value for money even when the capability gap stays open. The other story worth watching sits outside the top two entirely: SpaceX closed a sixty billion dollar all-stock acquisition of Cursor in mid-August, and Cursor's own blog confirmed the deal closed. I have not seen evidence that ownership change has touched the product I score here, so Cursor holds third on the same merits as before, IDE integration that still beats every agent above it. My order stays exactly where it was yesterday, and today's numbers give me clearer evidence for why that order holds.

Opus 5.5 Widens the Benchmark Gap

Claude Code with the new Opus 5.5 model scores 66 on Artificial Analysis's Coding Agent Index, four points ahead of Codex with GPT-6 Astra at 62.

Three New GPT-6 Tiers in One Month

OpenAI shipped Astra, Sol, and Luna between September 3 and 22, with Sol replacing GPT-5.5 at half its former promotional price.

SpaceX Closes the Cursor Deal

SpaceX's sixty billion dollar all-stock acquisition of Cursor closed in mid-August, though the product itself has not visibly changed since.

Order Holds at the Top

Claude Code keeps first, Codex keeps second, and Cursor keeps third on IDE integration strength that still leads the field.

2026-09-29

Claude Code holds first and OpenAI Codex holds second today, and the story I want to flag is a pricing shift that changes how I think about value for money across this whole table. Claude Code now ships bundled inside Claude's own subscription tiers, Pro at twenty dollars a month and Max starting at a hundred, a change from the old pay-per-token metering that made agentic coding sessions hard to budget. GitHub Copilot made a similar jump this year, switching to a credits system where one credit equals one cent, which lets Microsoft price a coding session the same way it prices any other consumption product. I like this shift for buyers, because it turns an agentic coding session into a flat monthly cost that is easy to plan around. The other story worth flagging sits lower in my table. Windsurf, ranked twelfth here, rebranded to Devin Desktop back in June after Cognition acquired the company, and windsurf.com now redirects straight to devin.ai. That leaves two names in my table, Windsurf and Devin, pointing at increasingly overlapping products under one owner, and I expect that overlap to force a rework of this list before the year is out. For now I keep both entries, because their current feature sets and pricing still differ enough to score on their own terms, Windsurf's low entry price against Devin's project based autonomous runs. Cursor holds third after SpaceX closed its sixty billion dollar acquisition of Anysphere in August, and its tiered credit system, free through two hundred dollars a month for Ultra, still gives it the widest spread of price points in the category.

Claude Code goes subscription-first

Claude Code now runs inside Claude Pro at twenty dollars a month and Max plans starting at a hundred, replacing the old pay-per-token billing that made long agent sessions hard to budget.

Copilot's credit system

GitHub Copilot switched to AI Credits priced at one cent each this year, letting Microsoft meter a coding session the same way it meters any other consumption product.

Windsurf's identity shift

Windsurf, ranked twelfth in my table, rebranded to Devin Desktop in June after Cognition's acquisition, and windsurf.com now redirects to devin.ai, a sign this list may need to merge the two entries eventually.

Cursor under new ownership

SpaceX closed its sixty billion dollar acquisition of Anysphere, Cursor's maker, in August, and Cursor's credit tiers still range from free to two hundred dollars a month for Ultra.

2026-09-28

Claude Code holds first and OpenAI Codex holds second, and the news I want to flag this week has nothing to do with a new model, it is about who owns the company behind rank three. SpaceX closed its acquisition of Cursor for sixty billion dollars in mid-August, and the deal is still reshaping how buyers think about seat costs. Cursor's Pro+ tier at sixty dollars a month now bundles Grok 4.6 access with triple the usage of the standard plan, and teams that used to spread three twenty-dollar Claude Pro seats across a squad for extra throughput are comparing that same sixty dollar spend against one Cursor Pro+ seat that keeps everything under a single harness. I still rank Claude Code first because its code quality and context window scores lead the table, and a team doing serious agentic refactoring gets more reliable diffs from it than from any competitor right now. Codex holds second on the strength of its own agentic runs, and I increasingly see both tools running side by side on the same project. Cursor earns its third place on IDE integration, the tab complete and multi file editing experience is still the smoothest of the three, and the SpaceX ownership change has not touched the product itself yet. What has changed is the buying conversation. If your team already juggles three or more seats to get around usage caps, this is the week to run the actual math on consolidating into one Cursor Pro+ subscription and dropping the separate seats. The sixty dollar price point is identical either way, the difference is whether you want one harness or the flexibility of picking a model per task.

SpaceX now owns Cursor

SpaceX closed a sixty billion dollar acquisition of Cursor in mid-August 2026, and the deal is reshaping how teams compare seat pricing.

Pro+ seat math got competitive

Cursor Pro+ at sixty dollars a month now bundles Grok 4.6 with triple usage, going head to head with three stacked twenty-dollar Claude Pro seats.

Claude Code still leads code quality

Its code quality and context window scores remain the highest in the table, the edge that keeps it first for serious agentic refactoring.

Codex and Claude Code run side by side

More teams now run both tools on the same project at once, without settling on a single default.

2026-09-27

Claude Code holds first and OpenAI Codex holds second, and this week the model underneath both tools moved. Release trackers report that Claude Opus 5.5 arrived on September 22, built for long-running agentic coding and knowledge work, and that OpenAI released GPT-6 Sol and GPT-6 Luna the same day with API prices cut in half. A coding assistant is only as good as the model behind it, so these releases matter to the ranking.

Claude Code stays first because its whole design is about long, multi-step tasks in a real repository. A model tuned for long-running agentic work fits that design closely, and when I hand a tool a refactor across many files, staying coherent for the full job is the quality I value above everything. Codex holds second with a cheaper OpenAI model family behind it, and lower API prices help anyone running many agent sessions in a day.

Cursor in third keeps its edge as an editor experience, with inline edits and fast tab completion that feel natural during everyday work. Google Antigravity at fourth and Google Jules at fifth benefit from Google's steady model cadence, and The Hacker News reports that Google also launched Gemini 3.8 Flash Cyber, a model aimed at vulnerability discovery, which shows how quickly the lab is specializing.

My advice is to match the tool to the job. Choose Claude Code for large, careful changes, Codex when cost per session matters, and Cursor when you live inside an editor. Every rank stays put today, because the new models strengthen the current top two without disturbing the rest.

Opus 5.5 targets long-running coding

Released September 22 for long-running agentic coding, which matches Claude Code's design.

GPT-6 Sol and Luna cut API prices in half

Trackers say OpenAI halved API prices on September 22, helping Codex users who run many sessions.

Cursor stays the editor favorite

Inline edits and fast completion keep it third.

Google specializes

The Hacker News reports Gemini 3.8 Flash Cyber for vulnerability discovery.

2026-09-26

Claude Code holds first and OpenAI Codex holds second, and this week I want to talk about release cadence, since a coding tool lives or dies by how quickly it improves. Releasebot's September log for Claude Code lists a steady run of fixes for MCP configuration, prompt history recall, and general stability. Those are the small changes I feel in daily work, and they explain why I keep it open all day.

Claude Code posts 9.8 for code quality, 9.7 for codebase understanding, and 9.7 for agentic capability. That combination is why I hand it the larger refactors and the tasks that touch many files at once.

Codex in second has a 9.6 agentic score, and it is the tool I use when I want a task delegated and reviewed afterward. Cursor in third has the best IDE score at 9.6, and it remains the friendliest place to work when I want to stay inside an editor and steer.

The New Stack's recent piece on these tools converging into a single stack matches my routine. I plan and refactor in Claude Code, delegate in Codex, and edit in Cursor. Ranks stay unchanged today because none of the reports moves a score I track. If you have to choose one, pick Claude Code for depth, Cursor for editor comfort, and Codex for delegation.

Claude Code leads quality

9.8 code quality and 9.7 codebase understanding suit big refactors.

Codex delegates well

A 9.6 agentic score fits hand-off tasks reviewed afterward.

Cursor owns the editor

Its 9.6 IDE score makes it the most comfortable place to steer.

Steady release cadence

Releasebot lists frequent Claude Code fixes across MCP and stability.

2026-09-25

Claude Code holds the top spot and OpenAI Codex holds second, and the most useful story this week is how developers are actually combining them. The New Stack published a piece arguing that Cursor, Claude Code, and Codex are converging into one stack that nobody planned, and that matches what I see in my own workflow. I use an editor-based tool for fast inline edits, then hand long multi-file tasks to a terminal agent.

The numbers still favor the two leaders. On Terminal-Bench 4.0, Codex with GPT-6 Astra scored 58.2 percent and Claude Code with Claude Fable 5.1 scored 57.9 percent. That is a gap of 0.3 points, which is inside normal run-to-run noise. I keep Claude Code first because of how it handles large repositories and long sessions, which is where my real work happens.

Cursor sits third, and a September leaderboard write-up credits its rise to the SpaceX acquisition completed August 14 and to Grok 4.7 shipping inside the editor on September 22. I could not open that page to confirm the details, so treat those specifics as reported and unverified. What I can say from use is that Cursor's multi-agent interface remains the smoothest way to run parallel tasks.

My advice is to pick one terminal agent and one editor, and learn both well. Rankings stay unchanged today, since nothing this week moves the benchmark picture.

Terminal-Bench 4.0

Codex scored 58.2 percent and Claude Code 57.9 percent, a 0.3 point gap.

Converging stack

The New Stack describes Cursor, Claude Code, and Codex being used together in one workflow.

Cursor news, reported

A leaderboard write-up cites the SpaceX acquisition and Grok 4.7 in Cursor; I could not verify the page.

Ranking unchanged

No benchmark movement this week, so scores stay put.