Best AI model overall right now
Claude Opus 5.5 leads the October 2026 index at 86.6, with the highest agentic (90) and knowledge (90) scores of any model here. GPT-6 Astra and GPT-6.1 Sol sit 2.3 points behind — and GPT-6.1 Sol does it at $10 per million output tokens against $50 for GPT-6 Astra. At the top, the difference between first and fifth is smaller than the difference between workloads.
Best AI model for coding in 2026
On the coding category, Claude Sonnet 5.5 scores highest among current models (85), ahead of Claude Opus 5.5 (83). Coding benchmarks reward getting a patch right; real engineering also needs the model to plan, run tools and recover from failures, which is what the agentic score measures. Pick the Coding preset to rank on both at once.
Best AI for agents and computer use
Claude Opus 5.5 has the strongest agentic score (90). Among the cheaper options, GPT-6.1 Sol scores 77 — worth testing for high-volume agents where per-step cost compounds.
Claude vs GPT vs Gemini: how the big three compare
| Best current model | Overall | Coding | Reasoning | $ in / out |
|---|---|---|---|---|
| Claude Opus 5.5 · Anthropic | 86.6 | 83 | 86 | $4 / $20 |
| GPT-6 Astra · OpenAI | 84.3 | 75 | 93 | $10 / $50 |
| Gemini 4 Argon · Google | 82.7 | 73 | 87 | $2 / $10 |
Anthropic currently holds the overall lead and the top coding score. OpenAI’s line leads on reasoning. Google’s Gemini 4 Argon, released October 13, lands within a few points of both at the low end of frontier pricing.
Best open-source (open-weight) LLMs
Open-weight models can be downloaded and self-hosted, which matters for data residency and per-token cost at scale. The strongest here:
- MiMo-V2.6-Pro · Xiaomi73.6
- Qwen3.8 Max · Alibaba69.9
- Step 5 Preview · StepFun69.3
- GLM-5.3 · Z.AI68.2
- DeepSeek V4.1 Flash · DeepSeek65.9
Cheapest frontier AI models and best value
A model is on the value frontier when nothing else is both better and cheaper. 4 models qualify: Claude Opus 5.5 ($8), GPT-6.1 Sol ($4), GPT-6.1 Terra ($2), MiMo-V2.6-Pro ($0.54). Every other model is beaten on both axes by at least one of these. Switch the explorer to Price vs performance to see the line, or Cost calculator to price your own volume.
AI coding tools built on these models
These are products, not models, so they are listed rather than ranked against the table.
CursorAI code editor · Anysphere
A VS Code-based editor built around multi-file edits and fast in-IDE iteration.
Runs on: Composer 2.5 (Cursor’s own model, built on Kimi K2.5) plus Claude, GPT and Gemini models.
Composer 2.5 (May 2026): 69.3% Terminal-Bench 2.0, 79.8% SWE-Bench Multilingual — Cursor-reported.
Claude CodeTerminal coding agent · Anthropic
An agent that works in your terminal and repository: reads code, runs commands, edits files and opens pull requests.
Runs on: Claude models, including Claude Opus 5.5 and Sonnet 5.5.
OpenAI CodexCloud and CLI coding agent · OpenAI
OpenAI’s software-engineering agent, available in ChatGPT, as a CLI and in IDE extensions.
Runs on: OpenAI GPT models.
New this month
Gemini 4 Argon from Google enters at #5 with 82.7 overall. Released to developers October 13, 2026. $2 / $10 is Google’s introductory rate through November 30, 2026.
GPT-6.1 Terra from OpenAI enters at #10 with 75.9 overall. Released October 2026; the lowest-priced model in the frontier top 10.
AI benchmark categories, explained
The overall score blends five kinds of tests. What each one measures, and which model leads it today:
- Agentic
- Multi-step tool use, computer use and long-running autonomous tasks.
- Leader: Claude Opus 5.5 (90)
- Coding
- Real-repository software engineering and code generation benchmarks.
- Leader: Claude Sonnet 5.5 (85)
Every model at a glance
Benchmarks, price and context window for each model on the leaderboard, in rank order.
Claude Opus 5.5 benchmarks and price#1
Claude Opus 5.5 by Anthropic scores 86.6 overall, strongest on multimodal (92). API pricing is $4 per million input tokens and $20 per million output tokens, with a 1M-token context window.
GPT-6 Astra benchmarks and price#2
GPT-6 Astra by OpenAI scores 84.3 overall, strongest on reasoning (93). API pricing is $10 per million input tokens and $50 per million output tokens, with a 1.05M-token context window.
GPT-6.1 Sol benchmarks and price#3
GPT-6.1 Sol by OpenAI scores 84.3 overall, strongest on reasoning (91). API pricing is $2 per million input tokens and $10 per million output tokens, with a 1.05M-token context window.
Claude Fable 5.1 benchmarks and price#4
Claude Fable 5.1 by Anthropic scores 83.1 overall, strongest on knowledge (86). API pricing is $10 per million input tokens and $50 per million output tokens, with a 1M-token context window.
Gemini 4 Argon benchmarks and price#5
Gemini 4 Argon by Google scores 82.7 overall, strongest on reasoning (87). API pricing is $2 per million input tokens and $10 per million output tokens, with a 1M-token context window. Released to developers October 13, 2026. $2 / $10 is Google’s introductory rate through November 30, 2026.
Claude Sonnet 5.5 benchmarks and price#6
Claude Sonnet 5.5 by Anthropic scores 82.2 overall, strongest on coding (85). API pricing is $2 per million input tokens and $10 per million output tokens, with a 1M-token context window.
Claude Fable 5 benchmarks and price#7
Claude Fable 5 by Anthropic scores 81.2 overall, strongest on knowledge (82). API pricing is $10 per million input tokens and $50 per million output tokens, with a 1M+-token context window. Anthropic has shipped a successor.
Claude Opus 5 benchmarks and price#8
Claude Opus 5 by Anthropic scores 79.9 overall, strongest on multimodal (91). API pricing is $5 per million input tokens and $25 per million output tokens. Anthropic has shipped a successor.
GPT-6 Sol benchmarks and price#9
GPT-6 Sol by OpenAI scores 78.8 overall, strongest on knowledge (79). API pricing is $2 per million input tokens and $10 per million output tokens, with a 1.05M-token context window. OpenAI has shipped a successor.
GPT-6.1 Terra benchmarks and price#10
GPT-6.1 Terra by OpenAI scores 75.9 overall, strongest on multimodal (92). API pricing is $1 per million input tokens and $5 per million output tokens, with a 1.05M-token context window. Its position is estimated from fewer benchmark results. Released October 2026; the lowest-priced model in the frontier top 10.
GPT-5.6 Sol benchmarks and price#11
GPT-5.6 Sol by OpenAI scores 75.1 overall, strongest on multimodal (90). API pricing is $4 per million input tokens and $20 per million output tokens, with a 1.05M-token context window. OpenAI has shipped a successor.
Muse Wonder Full benchmarks and price#12
Muse Wonder Full by Meta scores 74.9 overall, strongest on knowledge (78). API pricing is not listed on the index, with a 1M-token context window. Its position is estimated from fewer benchmark results.
MiMo-V2.6-Pro benchmarks and price#13
MiMo-V2.6-Pro by Xiaomi scores 73.6 overall, strongest on knowledge (69). API pricing is $0.43 per million input tokens and $0.87 per million output tokens, with a 1M-token context window. Weights are open, so it can be self-hosted. Its position is estimated from fewer benchmark results. Highest-ranked open-weight model on the index.
GPT-5.5 Pro benchmarks and price#14
GPT-5.5 Pro by OpenAI scores 73.4 overall. API pricing is $30 per million input tokens and $180 per million output tokens, with a 1M-token context window. Its position is estimated from fewer benchmark results.
Muse Wonder benchmarks and price#15
Muse Wonder by Meta scores 71.7 overall, strongest on knowledge (74). API pricing is $4 per million input tokens and $20 per million output tokens, with a 1M-token context window. Its position is estimated from fewer benchmark results.
Kimi K3 benchmarks and price#16
Kimi K3 by Moonshot AI scores 70.1 overall, strongest on multimodal (90). API pricing is $3 per million input tokens and $15 per million output tokens, with a 1.05M-token context window. Moonshot AI has shipped a successor. Succeeded by Kimi K3.1 Pro (October 23, 2026), which the index has not yet scored overall.
GPT-5.6 Terra benchmarks and price#17
GPT-5.6 Terra by OpenAI scores 70.0 overall, strongest on multimodal (85). API pricing is $2 per million input tokens and $12 per million output tokens, with a 1.05M-token context window. OpenAI has shipped a successor.
Qwen3.8 Max benchmarks and price#18
Qwen3.8 Max by Alibaba scores 69.9 overall, strongest on multimodal (92). API pricing is not listed on the index, with a 1M-token context window. Weights are open, so it can be self-hosted.
Claude Opus 4.8 benchmarks and price#19
Claude Opus 4.8 by Anthropic scores 69.7 overall, strongest on multimodal (88). API pricing is $5 per million input tokens and $25 per million output tokens, with a 1M-token context window. Anthropic has shipped a successor.
Step 5 Preview benchmarks and price#20
Step 5 Preview by StepFun scores 69.3 overall, strongest on multimodal (73). API pricing is $1 per million input tokens and $2.70 per million output tokens, with a 1M-token context window. Weights are open, so it can be self-hosted. Its position is estimated from fewer benchmark results.
GPT-5.5 benchmarks and price#21
GPT-5.5 by OpenAI scores 69.0 overall, strongest on multimodal (74). API pricing is $5 per million input tokens and $30 per million output tokens, with a 1M-token context window. OpenAI has shipped a successor.
Gemini 3.8 Flash benchmarks and price#22
Gemini 3.8 Flash by Google scores 68.8 overall, strongest on multimodal (88). API pricing is $0.75 per million input tokens and $3.75 per million output tokens, with a 1M-token context window.
GPT-5.4 Pro benchmarks and price#23
GPT-5.4 Pro by OpenAI scores 68.7 overall. API pricing is $30 per million input tokens and $180 per million output tokens, with a 1.05M-token context window. Its position is estimated from fewer benchmark results.
Claude Opus 4.7 (Adaptive) benchmarks and price#24
Claude Opus 4.7 (Adaptive) by Anthropic scores 68.6 overall, strongest on knowledge (65). API pricing is $5 per million input tokens and $25 per million output tokens, with a 1M-token context window. Anthropic has shipped a successor. Its position is estimated from fewer benchmark results.
GLM-5.3 benchmarks and price#25
GLM-5.3 by Z.AI scores 68.2 overall, strongest on reasoning (80). API pricing is not listed on the index, with a 1M-token context window. Weights are open, so it can be self-hosted. 753B parameters, open weights — self-hostable.
Grok 4.3 benchmarks and price#26
Grok 4.3 by xAI scores 68.0 overall. API pricing is $1.25 per million input tokens and $2.50 per million output tokens, with a 1M-token context window. Leads the index’s instruction-following board (97.1%).
DeepSeek V4.1 Flash benchmarks and price#27
DeepSeek V4.1 Flash by DeepSeek scores 65.9 overall. API pricing is not listed on the index, with a 1M-token context window. Weights are open, so it can be self-hosted.
How this leaderboard works
- One source for every score. All model numbers come from BenchLM as of October 26, 2026. Vendor launch charts use different harnesses, so mixing them makes a table look precise while ranking nothing.
- Gaps stay gaps. Where the index has not scored a model on a category, the cell shows “—”. Custom weights score each model only on what was measured and show the coverage.
- Evidence labels. “Estimated” means the index scored the model on fewer benchmarks, with a wider uncertainty range. “Superseded” means its lab has shipped a successor.
- Prices are list API rates per million tokens. “$ / 1M” blends input and output 3:1, the shape of typical chat traffic.
Frequently asked questions
What is the best AI model right now (October 2026)?
Claude Opus 5.5 from Anthropic leads with an overall score of 86.6, ahead of GPT-6 Astra (84.3) and GPT-6.1 Sol (84.3). The gap at the top is a few points, so the best pick depends on your workload — use the presets above to rank by coding, agents or research.
What is the best AI model for coding?
Claude Sonnet 5.5 has the highest coding score among current models (85). For autonomous, multi-step engineering work, also weigh the agentic score, where Claude Opus 5.5 leads (90).
Which is better: Claude, GPT or Gemini?
On this index the strongest current model from each lab is Claude Opus 5.5 (86.6), GPT-6 Astra (84.3), Gemini 4 Argon (82.7). Claude Opus 5.5 leads overall; GPT-6 Astra is strongest on reasoning among the three; Gemini 4 Argon matches the cheaper end of pricing at $2 / $10 per million tokens.
What is the best open-source (open-weight) LLM?
MiMo-V2.6-Pro by Xiaomi is the highest-ranked open-weight model, scoring 73.6. 5 open-weight models appear in this leaderboard; tick “Open weights only” to see them alone.
What is the cheapest frontier AI model?
GPT-6.1 Sol is the best-scoring model on the price-performance frontier at or under $5 per million blended tokens ($4). The “Price vs performance” view plots every model so you can see the trade-off directly.
When was Gemini 4 Argon released and how does it rank?
Google released Gemini 4 Argon to developers on October 13, 2026, after announcing it on September 30. It ranks in the top five overall here, with a 1M-token context window and an introductory price of $2 input / $10 output per million tokens through November 30, 2026.
Where do these benchmark scores come from?
All model scores come from one independent index, BenchLM, as of October 26, 2026, so every row is measured the same way. Each model’s details link to its source page. We do not mix vendor-reported numbers from different launch posts, because cross-lab methodologies differ.
How often is the AI leaderboard updated?
We re-check the index and refresh the table when frontier models launch or scores change materially. The date at the top of the page is the date of the data, not the date you loaded it.