benchr
AI model rankings
26 models across the frontier, mid, and open-weight tiers. Ranked by capability by default — flip the “Rank by” toggle to a Value lens that factors in price. Capability is benchr's synthesis of coding, reasoning, and writing, not an independent lab score. Sort any column, filter by type, license, or provider country, and open any model for the full review.
| # | Model | benchr Rating | SWE-bench % | Input $/1M | Output $/1M | Context | Tok/s | Released |
|---|---|---|---|---|---|---|---|---|
| 1 | Claude Fable 5Anthropicfrontier | 9.8 | 91.0% | $10.00 | $50.00 | 1M | 58 tok/s | Jun 2026 |
| 2 | Claude Sonnet 5Anthropicfrontier | 9.5 | 89.4% | $2.00 | $10.00 | 1M | 75 tok/s | Jul 2026 |
| 3 | Claude Opus 4.8Anthropicfrontier | 9.5 | 88.6% | $5.00 | $25.00 | 1M | 68 tok/s | May 2026 |
| 4 | Claude Opus 4.7Anthropicfrontier | 9.5 | 87.6% | $5.00 | $25.00 | 1M | 65 tok/s | Apr 2026 |
| 5 | GPT-5.6OpenAIfrontier | 9.4 | 89.8% | $5.00 | $30.00 | 1.1M | 85 tok/s | Jul 2026 |
| 6 | GPT-5.5OpenAIfrontier | 9.3 | 84.0% | $5.00 | $30.00 | 1.1M | 82 tok/s | Apr 2026 |
| 7 | DeepSeek V4-ProDeepSeekfrontier open | 9.1 | 80.6% | $0.435 | $0.870 | 1M | 95 tok/s | Apr 2026 |
| 8 | Grok 4.5xAIfrontier | 9.1 | 84.0% | $2.00 | $6.00 | 500K | 135 tok/s | Jul 2026 |
| 9 | GPT-5.4OpenAIfrontier | 9.0 | 80.0% | $2.50 | $15.00 | 1M | 84 tok/s | Mar 2026 |
| 10 | Claude Sonnet 4.6Anthropicmid | 8.8 | 79.6% | $3.00 | $15.00 | 1M | 95 tok/s | Feb 2026 |
| 11 | GPT-5OpenAIfrontier | 8.8 | 74.9% | $1.25 | $10.00 | 400K | 90 tok/s | Aug 2025 |
| 12 | GLM-5.2Z.AIfrontier open | 8.8 | 83.0% | $1.40 | $4.40 | 1M | 95 tok/s | Jun 2026 |
| 13 | MiniMax M3MiniMaxfrontier open | 8.7 | 82.0% | $0.300 | $1.20 | 1M | 110 tok/s | Jun 2026 |
| 14 | Gemini 3.5 FlashGooglemid | 8.6 | 80.6% | $1.50 | $9.00 | 1.0M | 289 tok/s | May 2026 |
| 15 | Gemini 3.1 ProGooglefrontier | 8.6 | 80.6% | $2.00 | $12.00 | 1M | 108 tok/s | Feb 2026 |
| 16 | Mistral Medium 3.5Mistralfrontier open | 8.6 | 77.6% | $1.50 | $7.50 | 256K | 105 tok/s | Apr 2026 |
| 17 | Qwen3.6-27BAlibabaopen | 8.6 | 77.2% | Free | Free | 262K | 105 tok/s | Apr 2026 |
| 18 | DeepSeek V4-FlashDeepSeekopen | 8.5 | 79.0% | $0.140 | $0.280 | 1M | 135 tok/s | Apr 2026 |
| 19 | Grok 4.3xAIfrontier | 8.2 | 68.0% | $1.25 | $2.50 | 1M | 120 tok/s | Apr 2026 |
| 20 | Llama 4 MaverickMetafrontier open | 8.0 | 66.0% | Free | Free | 1M | 120 tok/s | Apr 2025 |
| 21 | Kimi K2.6Moonshot AIfrontier open | 7.8 | 80.2% | $0.950 | $4.00 | 262K | 100 tok/s | Apr 2026 |
| 22 | Mistral Large 3Mistralfrontier open | 7.8 | 62.0% | $0.500 | $1.50 | 256K | 115 tok/s | Dec 2025 |
| 23 | Claude Haiku 4.5Anthropicsmall | 7.6 | 73.3% | $1.00 | $5.00 | 200K | 145 tok/s | Oct 2025 |
| 24 | Phi-4Microsoftsmall open | 7.4 | 30.0% | Free | Free | 16K | 220 tok/s | Dec 2024 |
| 25 | GPT-5 MiniOpenAIsmall | 7.3 | 48.0% | $0.250 | $2.00 | 400K | 160 tok/s | Aug 2025 |
| 26 | Llama 4 ScoutMetaopen | 7.3 | 56.0% | Free | Free | 10M | 180 tok/s | Apr 2025 |
How the benchr Rating works
The benchr Rating shows one number, and you choose what it means with the Rank by toggle. By default it's pure capability — what the model can do, independent of price. Switch to Value and price is folded in, rewarding cheap and free APIs. It's not a poll and no provider has paid for their position; the capability half is benchr's own editorial read, so treat it as an opinion with its math shown, not a lab measurement.
Quality — the default
Pure capability, the same composite the rest of the site uses. Built from benchmark results and the public record — inspectable in models.json, not from marketing materials.
Value — the optional lens
Capability blended with price efficiency, for "most capability per dollar." Price is the blended API rate (average of input and output per million tokens); free and self-hosted models score 100, scaling from $0.50 (full) down to $30.00 (zero).
Both run in assets/js/models.js, so you can read and verify them yourself. Each produces a 0–100 value shown on a 0–10 scale, and price is shown directly in the Input/Output columns regardless of mode. For verified official pricing and benchmark figures, see model-figures.json.
Methodology as of June 1, 2026. Formula may be revised as the model landscape changes — check the changelog for updates.
Frequently asked questions
What is the benchr Rating?
By default it's pure capability (coding, reasoning, writing). A “Rank by” toggle switches it to a Value lens that also folds in price efficiency (most capability per dollar). Both run in open JavaScript you can read in assets/js/models.js; price is shown in its own columns too.
Which AI model is best in 2026?
Depends on your budget and task. For no-budget-limit capability: Claude Opus 4.8. For the best capability-per-dollar: DeepSeek V4-Pro or Gemini 3.5 Flash. For free self-hosted: Llama 4 Maverick. Use the model recommender to get a personalized pick.
Are rankings ever paid or sponsored?
No. Rankings are computed purely from data in models.json. No provider has paid for placement. See editorial standards for the full policy.
How often is the data updated?
The goal is same-day updates when major models ship or providers change prices. The updated field in models.json shows the last data refresh. Spot an error? File a correction
Contact Privacy
Terms.
Why are benchmark scores labeled as "editorial estimates"?
Many benchmark figures aren't comparable across providers — test sets differ, conditions differ, and some numbers are self-reported. benchr's capability scores are built from available benchmark data but treated as estimates rather than certified figures. The methodology page explains the process. For verified official figures, see model-figures.json.