AI coding model leaderboard
Every coding model that matters, with its SWE-bench Verified score from one independent harness, API prices and context windows straight from each lab.
Updated yesterday, 2026-11-07. 33 models tracked. Sources: SWE-bench Verified (Vals.ai runs), Aider Polyglot, Artificial Analysis
SWE-bench Verified was retired as saturated on 1 September 2026, and labs stopped reporting it. Models released since then show no score here: the ranking reflects the benchmark, not which model is newest. Scores come from Vals.ai, where every model runs in the same agent setup.
- Best overall, top SWE-bench Verified
- Claude Opus 5Anthropic, 97.0% SWE-bench
- Best value, 70%+ at lowest output $
- GPT-5.6 LunaOpenAI, 93.0% SWE-bench
- Newest flagship, released 2026-10-26
- GPT-6.1 TerraOpenAI
All models
| # | Model | SWE-bench | Aider | Context | $/1M in | $/1M out | TPS | Tools | |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 1.Claude Opus 5 Anthropic, released 2026-07-24. Reasoning Top score on the Vals.ai SWE-bench Verified run; superseded by Opus 5.5, which was never tested on it. |
Claude Code | 97.0% | — | 1M | $5.0 | $25 | — | Claude Code |
| 2 | 2.DeepSeek V4 Pro (0813) DeepSeek, released 2026-08-13. Open weights, Reasoning Open weights and within a point of the top score; peak-hour API price shown, off-peak is half. |
| 96.4% | — | 1M | $1.3 | $4.0 | — | |
| 3 | 3.GPT-5.6 Sol OpenAI, released 2026-07-09. Reasoning Price shown is promotional until 2026-11-30; input over 272K tokens costs more. |
| 96.2% | — | 1M | $4.0 | $20 | — | |
| 4 | 4.Grok 4.6 xAI, released 2026-08-12. Reasoning Low output price for its score; xAI states no fixed output limit. Prompts over 200K tokens cost double. |
| 95.6% | — | 500K | $2.0 | $6.0 | — | |
| 5 | 5.GPT-5.6 Terra OpenAI, released 2026-07-09. Reasoning Close to 5.6 Sol on SWE-bench Verified at half the input price. |
| 95.4% | — | 1M | $2.0 | $12 | — | |
| 6 | 6.Claude Fable 5 Anthropic, released 2026-06-09. Reasoning Superseded by Fable 5.1; the most expensive Claude model per token. |
Claude Code | 95.0% | — | 1M | $10 | $50 | — | Claude Code |
| 7 | 7.GPT-5.6 Luna OpenAI, released 2026-07-09. Reasoning Cheapest model here scoring above 90% on SWE-bench Verified. |
| 93.0% | — | 1M | $0.20 | $1.2 | — | |
| 8 | 8.Claude Opus 4.8 Anthropic, released 2026-05-28. Reasoning Same price as Opus 5 with a lower score; Vals measured 85.8% when run inside Claude Code. |
Claude Code | 88.6% | — | 1M | $5.0 | $25 | — | Claude Code |
| 9 | 9.Qwen 3.8 27B Alibaba, released 2026-08-05. Open weights, Reasoning Apache-2.0 open weights at 27B parameters; small enough to self-host on one large GPU node. |
| 86.0% | — | 262K | $0.50 | $3.0 | — | |
| 10 | 10.GPT-5.5 OpenAI, released 2026-04-24. Reasoning Priced above the newer 5.6 models, which score higher. |
| 82.6% | — | 1M | $5.0 | $30 | — | |
| 11 | 11.Claude Opus 4.7 Anthropic, released 2026-04-16. Reasoning Older Opus at the same price as Opus 4.8 and Opus 5, which score higher. |
Claude Code, Cursor, Windsurf, Zed | 82.0% | — | 1M | $5.0 | $25 | — | Claude Code, Cursor, Windsurf +1 |
| 12 | 12.Gemini 3.8 Flash Google, released 2026-09-02. Reasoning Newest Gemini Flash; price shown is introductory until 2026-12-31, then $1.50 / $7.50. |
| 80.0% | — | 1M | $0.75 | $3.8 | — | |
| 13 | 13.Claude Sonnet 5 Anthropic, released 2026-06-30. Reasoning Superseded by Sonnet 5.5 at the same price. |
Claude Code | 79.6% | — | 1M | $2.0 | $10 | — | Claude Code |
| 14 | 14.GPT-5 (high reasoning) OpenAI, released 2025-08-07. Reasoning Deprecated, API shutdown 2026-12-11; still the top Aider Polyglot score (88.0%). |
Cursor, Windsurf, GitHub Copilot, Bolt.new, v0 | — | 88.0% | 400K | $1.3 | $10 | — | Cursor, Windsurf, GitHub Copilot +3 |
| 15 | 15.Gemini 3.1 Pro (preview) Google, released 2026-02-19. Reasoning Still a preview model; prompts over 200K tokens cost $4 / $18. |
| 78.8% | — | 1M | $2.0 | $12 | — | |
| 16 | 16.GPT-5.4 (xhigh) OpenAI, released 2026-03-05. Reasoning Score is at xhigh reasoning effort; the newer 5.6 Luna scores higher for less. |
| 78.2% | — | 1M | $2.5 | $15 | — | |
| 17 | 17.Claude Opus 4.6 Anthropic, released 2026-02-05. Reasoning Older Opus; same price as newer Opus models that score higher. |
Claude Code, Cursor, Windsurf, Replit, Zed | 78.2% | — | 1M | $5.0 | $25 | — | Claude Code, Cursor, Windsurf +2 |
| 18 | 18.Claude Sonnet 4.6 Anthropic, released 2026-02-17. Reasoning Costs more than Sonnet 5 and Sonnet 5.5, which replace it. |
Claude Code, Cursor, Windsurf, Cline, Zed | 77.4% | — | 1M | $3.0 | $15 | — | Claude Code, Cursor, Windsurf +5 |
| 19 | 19.Claude Opus 4.5 Anthropic, released 2025-11-24. Reasoning Last Opus with a 200K context; Anthropic will not retire it before 2026-11-24. |
Claude Code, Cursor | 76.4% | — | 200K | $5.0 | $25 | — | Claude Code, Cursor |
| 20 | 20.GPT-5.2 OpenAI, released 2025-12-11. Reasoning Still offered, but GPT-5.6 Luna scores higher at a fraction of the price. |
Cursor, Windsurf, GitHub Copilot, Bolt.new, v0 | 75.8% | — | 400K | $1.8 | $14 | — | Cursor, Windsurf, GitHub Copilot +4 |
| 21 | 21.Gemini 3 Flash (preview) Google, released 2025-12-17. Reasoning Still offered as a preview; Google lists Gemini 3.6 Flash as its replacement. |
Cursor, Windsurf, Cline, Aider | 75.0% | — | 1M | $0.50 | $3.0 | — | Cursor, Windsurf, Cline +1 |
| 22 | 22.Claude Haiku 4.5 Anthropic, released 2025-10-15. Reasoning Score is with extended thinking on; Haiku 5.5 now costs less per token. |
Claude Code, Cursor, Zed | 66.6% | — | 200K | $1.0 | $5.0 | — | Claude Code, Cursor, Zed |
| 23 | 23.Claude Opus 5.5 Anthropic, released 2026-09-22. Reasoning Current Anthropic default; released after Vals archived SWE-bench Verified, so no score. |
Claude Code | — | — | 1M | $4.0 | $20 | — | Claude Code |
| 24 | 24.Claude Fable 5.1 Anthropic, released 2026-09-01. Reasoning Anthropic's model for the hardest long-horizon work; not tested on SWE-bench Verified. |
Claude Code | — | — | 1M | $10 | $50 | — | Claude Code |
| 25 | 25.Claude Sonnet 5.5 Anthropic, released 2026-09-28. Reasoning Half the per-token price of Opus 5.5; not tested on SWE-bench Verified. |
Claude Code | — | — | 1M | $2.0 | $10 | — | Claude Code |
| 26 | 26.Claude Haiku 5.5 Anthropic, released 2026-10-07. Reasoning Price shown is for prompts up to 100K tokens; longer prompts cost $0.50 / $2.50. |
Claude Code | — | — | 1M | $0.10 | $0.50 | — | Claude Code |
| 27 | 27.GPT-6 Astra OpenAI, released 2026-09-03. Reasoning OpenAI's current flagship; OpenAI no longer reports SWE-bench Verified. |
| — | — | 1M | $10 | $50 | — | |
| 28 | 28.GPT-6.1 Sol OpenAI, released 2026-09-29. Reasoning OpenAI pitches it as close to Astra at a fifth of the price; not tested on SWE-bench Verified. |
| — | — | 1M | $2.0 | $10 | — | |
| 29 | 29.GPT-6.1 Terra OpenAI, released 2026-10-26. Reasoning Cheaper tier of the 6.1 family; not tested on SWE-bench Verified. |
| — | — | 1M | $1.0 | $5.0 | — | |
| 30 | 30.Gemini 4 Argon Google, released 2026-10-13. Reasoning Google's current flagship; price shown runs until 2026-11-30, then $4 / $20. |
| — | — | 1M | $2.0 | $10 | — | |
| 31 | 31.Grok 4.7 xAI, released 2026-09-21. Reasoning Replaces Grok 4.6 at the same price; not tested on SWE-bench Verified. |
Cursor, Grok Build | — | — | 500K | $2.0 | $6.0 | — | Cursor, Grok Build |
| 32 | 32.DeepSeek V4.1 Flash DeepSeek, released 2026-09-10. Open weights, Reasoning Open weights; serves the deepseek-flash API id (updated 2026-10-17). Peak-hour price shown. |
| — | — | 1M | $0.30 | $1.2 | — | |
| 33 | 33.Qwen 3 Coder 480B Alibaba, released 2025-07-22. Open weights Open weights; not on Alibaba Cloud's international price list, so the price shown is unverified. |
Cline, Aider | — | — | 262K | $0.40 | $1.6 | — | Cline, Aider |
What the columns mean
- SWE-bench Verified
- 500 real GitHub issues the model has to solve end-to-end. 70%+ is production-viable.
- Aider Polyglot
- 225 Exercism exercises across C++, Go, Java, JavaScript, Python, Rust. Edit accuracy, not pattern-match.
- Context
- Max tokens the model reads per request. Bigger = can hold more of your codebase at once.
- TPS
- Output tokens per second. Reasoning models look slower because they burn tokens on internal thinking.
- Tools
- Products that actually expose this model. Changes often — missing names probably means the tool hasn't shipped it yet.
How we update this
We re-read Vals.ai's SWE-bench results and each lab's model and pricing pages when we update this page; the date at the top says when that last happened. If a score is "—", that benchmark hasn't tested the model yet; we'd rather say that than fill the cell with a guess.
Picked a model? Now pick the tool.
Most of these models are available across several tools. The tool shapes your workflow as much as the model shapes your output.
Coding model questions
- Which AI model is best for coding right now?
- Claude Opus 5 (Anthropic) has the highest SWE-bench Verified score here, 97%. The benchmark was retired in September 2026, so newer models such as GPT-6.1 Terra have no score; pick by price, context and the tools you use as well.
- What is SWE-bench Verified?
- A benchmark of real GitHub issues from popular Python projects, human-checked to be solvable. A model scores by producing a patch that passes the project's tests; the percentage is the share of issues it fixes.
- Which strong coding model is cheapest?
- Among models scoring 70% or more on SWE-bench Verified, GPT-5.6 Luna has the lowest output price here: $1.2 per million output tokens.
- How current is this leaderboard?
- Scores and prices were last checked on 2026-11-07, from Vals.ai's SWE-bench Verified runs (one agent setup for every model) and each lab's own model and pricing pages. A blank score means the benchmark has not tested that model.