Model data

Find the right model for the job

Choose a benchmark below. Higher dots score better; dots farther left cost less on the selected cost basis. Each dot is a model at one thinking level.

Explore by providerClick to focus · click again to show all
All providers · 78 models

The full comparison. Choose a provider above to explore its models.

General intelligence · Reported cost / task

About this benchmark

Intelligence Index

General intelligence

Artificial Analysis v4.3 · saved reasoning levels

What it tests

A composite of Artificial Analysis’s evaluation suite: reasoning, knowledge, coding and agentic work, run the same way for every model.

Use it as the broad first read: how capable a model is overall, and what it costs to get that capability, before narrowing to a coding-specific benchmark.

What the score tells you

The index runs from 0 to 100; higher is more capable across the suite. It is a composite score, not a pass rate or a chance that your task succeeds.

A broad average can hide weaknesses on the specific job you need. Each reasoning level is tested as its own configuration. Earlier generations the source has retired from its default list keep their measured results. Scores are Artificial Analysis’s own runs with its own prompts and harness.

Reported cost / task

Artificial Analysis’s published cost to run the index, divided by its tasks, at its listed prices. It publishes no token breakdown, so Lens cannot reprice it with its own table.

Methodology and published results ↗

AnthropicOpenAISpaceXAIGoogleMetaZ.aiMoonshotDeepSeekQwenSame model across thinking levelsClick a dot to focus its model

57.6 / 100

Highest tested score

Opus 5.5 · max · $5.98 · reported cost / task

$1.82

Cheapest within 5 points of the top

Opus 5.5 · high · 53.6 / 100 · a budget shortlist, not a proven tie

15 of 139

Worth comparing

No selected option scores higher for the same or lower cost, or costs less for the same score. Small gaps may be noise.

Intelligence Index methodology and results ↗ · 139 configurations shown · 0 source configurations excluded without usable scores or costs. Snapshot captured Sep 29, 2026. API prices checked Sep 28, 2026. Each benchmark uses its own workload and cost basis; compare costs only within a view.

Artificial Analysis Intelligence Index v4.3, captured Sep 29, 2026. Every reasoning level the source publishes is its own dot, including non-reasoning modes. Cost is the source’s published cost to run the index per task at its listed prices; it has no token breakdown, so Lens does not reprice it. Scores the source marks as estimated are left out. The index measures broad capability, not coding alone. Use the coding benchmarks to narrow the choice.

Your session cost vs DeepSWE score

Different workloads, shown for context rather than a controlled comparison: your average cost per session on the horizontal axis, against the pass rate DeepSWE measured for that model and reasoning effort on the vertical axis.

Import your sessions to price your own work

Your agents run longer tasks than a benchmark, or shorter ones, and lean on the cache differently. Import your logs and Lens will plot what an average session costs you on each model.

Advanced: combined Lens coding profile

Lens coding profile · v2

A score you can check

How Lens scores a model

Each score belongs to a model at one reasoning level. More thinking can improve its results, but can also cost more.

  1. Quality contributes up to 75 points. We combine engineering (40%), terminal work (30%), coding (15%), reasoning (10%), and following instructions (5%). Multiply that quality score by 0.75.
  2. Affordability contributes up to 25 points. Divide the cheapest configuration’s estimated cost by this configuration’s estimated cost, then multiply by 25. The cheapest gets all 25 points; one costing twice as much gets 12.5.
  3. Add the two contributions. This is the Lens score out of 100. These weights are our preference for coding work, not a measured probability of success.

The chart keeps quality and cost separate. A point is on the frontier when no other point is at least as good on both and better on one. That is different from ranking by the total Lens score.

DeepSWE · engineeringMeasured74.12 × 40% = 29.65
Terminal-Bench · terminal workMeasured57.88 × 30% = 17.36
LiveBench · codingEstimated81.33 × 15% = 12.20
LiveBench · reasoningEstimated93.77 × 10% = 9.38
LiveBench · instructionsEstimated76.49 × 5% = 3.82

Quality: 72.41 × 0.75 = 54.31 points

Affordability: $0.014 ÷ $4.43 × 25 = 0.08 points

Lens score 54.39 / 100

Displayed numbers are rounded; calculations use full precision. 70% of this configuration’s quality weight comes from directly measured results.

What if a benchmark has not tested this level?

Hollow dots include estimates. We use another tested level of the same model and scale its result by the change in its DeepSWE score. If there is no matching anchor, we use this level’s DeepSWE score as a proxy. This assumption has not been validated. “Measured only” removes these estimates.

Cost uses DeepSWE’s token usage and listed prices. It excludes tool and infrastructure costs and is not your bill. Benchmarks can use different agents. Scores and ranks can change as evidence or the comparison group changes.

Three benchmark sources, five quality checks. Rank uses 75% quality and 25% estimated DeepSWE cost. The frontier compares quality and cost directly, without the ranking weights.

69 scored configurations across 27 models · 60 provisional estimates. 1 model identities lack pricing.

Each available reasoning level receives its own score and DeepSWE token-cost estimate. Estimated components use a same-model anchor scaled by the DeepSWE effort ratio, or a DeepSWE-only proxy when no anchor exists. This transfer assumption is unvalidated; estimated scores can change the frontier.

Missing prices: qwen3.8-max

Compare tested model/agent configurations, not isolated model ability. Matching model and effort does not make different harnesses equivalent. This profile measures coding-related performance, not universal quality, speed, or your own success rate.

AnthropicOpenAISpaceXAIGoogleMetaZ.aiMoonshotDeepSeekSame model across thinking levelsClick a dot to focus its model

Hollow dots include estimated components; filled dots have all five measured components. Lines connect each model’s available reasoning levels. Quality is an explicitly weighted index, not a combined pass percentage. No composite confidence interval is available. Small differences do not establish significant superiority. Cost excludes tools, infrastructure and unreported billing dimensions; it is not measured cost across the full benchmark suite.

RankConfigurationLens score / 100Quality / 100EvidenceDeepSWE est. $ / attemptTerminal agentFrontier
1GPT-6 Astra · xhigh54.3972.41Estimated · 70%$4.43CodexYes
2GPT-6 Astra · max53.9371.84Measured · 100%$7.50CodexNo
3GPT-6 Astra · high53.9171.75Estimated · 70%$3.92CodexYes
4GPT-5.5 · xhigh53.8671.75Estimated · 70%$7.11No exact matchNo
5GPT-6 Astra · medium52.8770.33Estimated · 70%$3.08CodexYes
Sources, weights, and coverage

Snapshot checked September 22, 2026 UTC. Weights are Lens’s coding-workflow preference, not scientifically optimal weights. Equal means within LiveBench categories prevent categories with more subtasks from gaining extra weight.

  • DeepSWE v1.1 — 40% · repository engineering · mini-SWE-agent · 113 tasks
  • Terminal-Bench 4.0 — 30% · terminal workflows · published model/agent pairs; agents differ
  • LiveBench 2026-06-25 — 30% · coding 15%, reasoning 10%, instructions 5% · equal subtask means within each category

All reasoning levels includes measured and estimated scores; Measured only requires all five measured checks. Observed values always retain exact model and effort matches. For missing components, use the nearest measured effort of the same model, multiply its component score by target DeepSWE pass rate / anchor DeepSWE pass rate, and bound to 0–100. A zero-pass anchor is never used. Without an anchor, use the target DeepSWE score as an explicitly labeled proxy. Weights never redistribute. LiveBench rows with unspecified effort are excluded. Terminal agents vary by provider and are shown in the table. Scores describe those published systems.

Lens score = 0.75 × quality + 0.25 × (cheapest eligible DeepSWE cost ÷ this configuration’s DeepSWE cost × 100). Rank can change as the eligible cohort changes. Tokens are already reflected in cost and receive no separate bonus; no latency ranking is claimed.

LiveCodeBench’s downloaded default data ended in April 2025 and lacked current model coverage. SWE-Bench Pro was left out to avoid adding another overlapping repository benchmark with mixed harnesses. Neither contributes hidden scores.

ConfigurationDeepSWE · engineering (40%)Terminal-Bench · terminal work (30%)LiveBench · coding (15%)LiveBench · reasoning (10%)LiveBench · instructions (5%)Quality / 100Coverage by weightStatus
GPT-6 Astra · xhigh74.1257.8881.33Estimated: Estimated from gpt-6-astra · max: 80.36 × (74.12 / 73.23), bounded to 0–10093.77Estimated: Estimated from gpt-6-astra · max: 92.65 × (74.12 / 73.23), bounded to 0–10076.49Estimated: Estimated from gpt-6-astra · max: 75.58 × (74.12 / 73.23), bounded to 0–10072.4170%Provisional rank
GPT-6 Astra · maxLiveBench: gpt-6-astra-max73.2358.1880.3692.6575.5871.84100%Measured components
GPT-6 Astra · high73.2357.8880.36Estimated: Estimated from gpt-6-astra · max: 80.36 × (73.23 / 73.23), bounded to 0–10092.65Estimated: Estimated from gpt-6-astra · max: 92.65 × (73.23 / 73.23), bounded to 0–10075.58Estimated: Estimated from gpt-6-astra · max: 75.58 × (73.23 / 73.23), bounded to 0–10071.7570%Provisional rank
GPT-5.5 · xhighLiveBench: gpt-5.5-xhigh67.0467.04Estimated: DeepSWE-only proxy: no same-model measured anchor available82.1589.6570.7371.7570%Provisional rank
GPT-6 Astra · medium72.7954.2479.87Estimated: Estimated from gpt-6-astra · max: 80.36 × (72.79 / 73.23), bounded to 0–10092.09Estimated: Estimated from gpt-6-astra · max: 92.65 × (72.79 / 73.23), bounded to 0–10075.12Estimated: Estimated from gpt-6-astra · max: 75.58 × (72.79 / 73.23), bounded to 0–10070.3370%Provisional rank
Opus 5 · xhigh73.1553.9480.90Estimated: Estimated from claude-opus-5 · max: 81.45 × (73.15 / 73.65), bounded to 0–10090.60Estimated: Estimated from claude-opus-5 · max: 91.21 × (73.15 / 73.65), bounded to 0–10063.34Estimated: Estimated from claude-opus-5 · max: 63.77 × (73.15 / 73.65), bounded to 0–10069.8170%Provisional rank
Opus 5 · maxLiveBench: claude-opus-5-max-effort73.6551.8281.4591.2163.7769.53100%Measured components
GPT-5.5 · high64.3864.38Estimated: DeepSWE-only proxy: no same-model measured anchor available78.90Estimated: Estimated from gpt-5.5 · xhigh: 82.15 × (64.38 / 67.04), bounded to 0–10086.10Estimated: Estimated from gpt-5.5 · xhigh: 89.65 × (64.38 / 67.04), bounded to 0–10067.93Estimated: Estimated from gpt-5.5 · xhigh: 70.73 × (64.38 / 67.04), bounded to 0–10068.9140%Provisional rank
Kimi K3 · max68.5168.51Estimated: DeepSWE-only proxy: no same-model measured anchor available68.51Estimated: DeepSWE-only proxy: no same-model measured anchor available68.51Estimated: DeepSWE-only proxy: no same-model measured anchor available68.51Estimated: DeepSWE-only proxy: no same-model measured anchor available68.5140%Provisional rank
Opus 5 · high72.8350.3080.54Estimated: Estimated from claude-opus-5 · max: 81.45 × (72.83 / 73.65), bounded to 0–10090.20Estimated: Estimated from claude-opus-5 · max: 91.21 × (72.83 / 73.65), bounded to 0–10063.06Estimated: Estimated from claude-opus-5 · max: 63.77 × (72.83 / 73.65), bounded to 0–10068.4770%Provisional rank
Fable 5 · xhigh69.9144.67Estimated: Estimated from claude-fable-5 · max: 44.55 × (69.91 / 69.72), bounded to 0–10086.22Estimated: Estimated from claude-fable-5 · max: 85.99 × (69.91 / 69.72), bounded to 0–10089.89Estimated: Estimated from claude-fable-5 · max: 89.65 × (69.91 / 69.72), bounded to 0–10075.97Estimated: Estimated from claude-fable-5 · max: 75.77 × (69.91 / 69.72), bounded to 0–10067.0940%Provisional rank
Fable 5 · maxLiveBench: claude-fable-5-max-effort69.7244.5585.9989.6575.7766.91100%Measured components
Fable 5 · high68.6043.83Estimated: Estimated from claude-fable-5 · max: 44.55 × (68.60 / 69.72), bounded to 0–10084.61Estimated: Estimated from claude-fable-5 · max: 85.99 × (68.60 / 69.72), bounded to 0–10088.21Estimated: Estimated from claude-fable-5 · max: 89.65 × (68.60 / 69.72), bounded to 0–10074.55Estimated: Estimated from claude-fable-5 · max: 75.77 × (68.60 / 69.72), bounded to 0–10065.8340%Provisional rank
GPT-5.6 Sol · maxLiveBench: gpt-5.6-sol-max72.6737.2783.9491.6571.8565.60100%Measured components
GPT-6 Astra · low67.0450.6173.56Estimated: Estimated from gpt-6-astra · max: 80.36 × (67.04 / 73.23), bounded to 0–10084.82Estimated: Estimated from gpt-6-astra · max: 92.65 × (67.04 / 73.23), bounded to 0–10069.19Estimated: Estimated from gpt-6-astra · max: 75.58 × (67.04 / 73.23), bounded to 0–10064.9770%Provisional rank
Opus 5 · medium68.9044.8576.20Estimated: Estimated from claude-opus-5 · max: 81.45 × (68.90 / 73.65), bounded to 0–10085.34Estimated: Estimated from claude-opus-5 · max: 91.21 × (68.90 / 73.65), bounded to 0–10059.66Estimated: Estimated from claude-opus-5 · max: 63.77 × (68.90 / 73.65), bounded to 0–10063.9670%Provisional rank
GLM 5.3 Flash · max63.3963.39Estimated: DeepSWE-only proxy: no same-model measured anchor available63.39Estimated: DeepSWE-only proxy: no same-model measured anchor available63.39Estimated: DeepSWE-only proxy: no same-model measured anchor available63.39Estimated: DeepSWE-only proxy: no same-model measured anchor available63.3940%Provisional rank
GPT-5.6 Sol · xhigh70.7336.28Estimated: Estimated from gpt-5.6-sol · max: 37.27 × (70.73 / 72.67), bounded to 0–10081.71Estimated: Estimated from gpt-5.6-sol · max: 83.94 × (70.73 / 72.67), bounded to 0–10089.21Estimated: Estimated from gpt-5.6-sol · max: 91.65 × (70.73 / 72.67), bounded to 0–10069.93Estimated: Estimated from gpt-5.6-sol · max: 71.85 × (70.73 / 72.67), bounded to 0–10063.8540%Provisional rank
DeepSeek V4 Pro · max62.8362.83Estimated: DeepSWE-only proxy: no same-model measured anchor available62.83Estimated: DeepSWE-only proxy: no same-model measured anchor available62.83Estimated: DeepSWE-only proxy: no same-model measured anchor available62.83Estimated: DeepSWE-only proxy: no same-model measured anchor available62.8340%Provisional rank
Muse Spark 1.2 · xhighLiveBench: muse-spark-1.2-xhigh54.8754.87Estimated: DeepSWE-only proxy: no same-model measured anchor available77.5490.0074.3362.7670%Provisional rank
GPT-5.6 Sol · high69.4035.60Estimated: Estimated from gpt-5.6-sol · max: 37.27 × (69.40 / 72.67), bounded to 0–10080.17Estimated: Estimated from gpt-5.6-sol · max: 83.94 × (69.40 / 72.67), bounded to 0–10087.54Estimated: Estimated from gpt-5.6-sol · max: 91.65 × (69.40 / 72.67), bounded to 0–10068.62Estimated: Estimated from gpt-5.6-sol · max: 71.85 × (69.40 / 72.67), bounded to 0–10062.6540%Provisional rank
Fable 5 · medium65.3741.77Estimated: Estimated from claude-fable-5 · max: 44.55 × (65.37 / 69.72), bounded to 0–10080.62Estimated: Estimated from claude-fable-5 · max: 85.99 × (65.37 / 69.72), bounded to 0–10084.05Estimated: Estimated from claude-fable-5 · max: 89.65 × (65.37 / 69.72), bounded to 0–10071.04Estimated: Estimated from claude-fable-5 · max: 75.77 × (65.37 / 69.72), bounded to 0–10062.7340%Provisional rank
Muse Spark 1.1 · xhighLiveBench: muse-spark-1.1-xhigh53.3253.32Estimated: DeepSWE-only proxy: no same-model measured anchor available77.1687.7369.6461.1570%Provisional rank
GLM 5.3 · max68.9641.8268.96Estimated: DeepSWE-only proxy: no same-model measured anchor available68.96Estimated: DeepSWE-only proxy: no same-model measured anchor available68.96Estimated: DeepSWE-only proxy: no same-model measured anchor available60.8270%Provisional rank
GPT-5.4 · xhighLiveBench: gpt-5.4-xhigh51.7751.77Estimated: DeepSWE-only proxy: no same-model measured anchor available77.5488.1270.2260.1970%Provisional rank
Gemini 3.8 Flash · highLiveBench: gemini-3.8-flash-high73.8319.0972.4989.2981.4159.13100%Measured components
GPT-5.6 Terra · maxLiveBench: gpt-5.6-terra-max69.6221.5278.2590.6364.6258.34100%Measured components
GPT-5.5 · medium53.9853.98Estimated: DeepSWE-only proxy: no same-model measured anchor available66.15Estimated: Estimated from gpt-5.5 · xhigh: 82.15 × (53.98 / 67.04), bounded to 0–10072.20Estimated: Estimated from gpt-5.5 · xhigh: 89.65 × (53.98 / 67.04), bounded to 0–10056.96Estimated: Estimated from gpt-5.5 · xhigh: 70.73 × (53.98 / 67.04), bounded to 0–10057.7840%Provisional rank
Fable 5 · low59.5838.07Estimated: Estimated from claude-fable-5 · max: 44.55 × (59.58 / 69.72), bounded to 0–10073.49Estimated: Estimated from claude-fable-5 · max: 85.99 × (59.58 / 69.72), bounded to 0–10076.61Estimated: Estimated from claude-fable-5 · max: 89.65 × (59.58 / 69.72), bounded to 0–10064.75Estimated: Estimated from claude-fable-5 · max: 75.77 × (59.58 / 69.72), bounded to 0–10057.1840%Provisional rank
Gemini 3.8 Flash · medium71.0218.36Estimated: Estimated from gemini-3.8-flash · high: 19.09 × (71.02 / 73.83), bounded to 0–10069.73Estimated: Estimated from gemini-3.8-flash · high: 72.49 × (71.02 / 73.83), bounded to 0–10085.90Estimated: Estimated from gemini-3.8-flash · high: 89.29 × (71.02 / 73.83), bounded to 0–10078.32Estimated: Estimated from gemini-3.8-flash · high: 81.41 × (71.02 / 73.83), bounded to 0–10056.8840%Provisional rank
GPT-5.6 Luna · maxLiveBench: gpt-5.6-luna-max67.1917.2782.9185.6460.1256.06100%Measured components
Gemini 3.6 Flash · highLiveBench: gemini-3.6-flash-high46.6846.68Estimated: DeepSWE-only proxy: no same-model measured anchor available77.8685.1575.3756.6470%Provisional rank
Opus 4.8 · maxLiveBench: claude-opus-4-8-max-effort58.9723.6481.8389.1972.0355.48100%Measured components
GPT-5.6 Sol · medium61.0631.32Estimated: Estimated from gpt-5.6-sol · max: 37.27 × (61.06 / 72.67), bounded to 0–10070.54Estimated: Estimated from gpt-5.6-sol · max: 83.94 × (61.06 / 72.67), bounded to 0–10077.02Estimated: Estimated from gpt-5.6-sol · max: 91.65 × (61.06 / 72.67), bounded to 0–10060.37Estimated: Estimated from gpt-5.6-sol · max: 71.85 × (61.06 / 72.67), bounded to 0–10055.1240%Provisional rank
DeepSeek V4 Flash · max53.3253.32Estimated: DeepSWE-only proxy: no same-model measured anchor available53.32Estimated: DeepSWE-only proxy: no same-model measured anchor available53.32Estimated: DeepSWE-only proxy: no same-model measured anchor available53.32Estimated: DeepSWE-only proxy: no same-model measured anchor available53.3240%Provisional rank
Gemini 3.7 Flash · medium65.4911.25Estimated: Estimated from gemini-3.7-flash · high: 11.21 × (65.49 / 65.27), bounded to 0–10079.16Estimated: Estimated from gemini-3.7-flash · high: 78.89 × (65.49 / 65.27), bounded to 0–10088.10Estimated: Estimated from gemini-3.7-flash · high: 87.80 × (65.49 / 65.27), bounded to 0–10080.20Estimated: Estimated from gemini-3.7-flash · high: 79.93 × (65.49 / 65.27), bounded to 0–10054.2640%Provisional rank
Gemini 3.7 Flash · highLiveBench: gemini-3.7-flash-high65.2711.2178.8987.8079.9354.08100%Measured components
Grok 4.6 · medium67.4821.01Estimated: Estimated from grok-4.6 · high: 20.30 × (67.48 / 65.19), bounded to 0–10067.48Estimated: DeepSWE-only proxy: no same-model measured anchor available67.48Estimated: DeepSWE-only proxy: no same-model measured anchor available67.48Estimated: DeepSWE-only proxy: no same-model measured anchor available53.5440%Provisional rank
Opus 5 · low58.1334.8564.28Estimated: Estimated from claude-opus-5 · max: 81.45 × (58.13 / 73.65), bounded to 0–10071.99Estimated: Estimated from claude-opus-5 · max: 91.21 × (58.13 / 73.65), bounded to 0–10050.33Estimated: Estimated from claude-opus-5 · max: 63.77 × (58.13 / 73.65), bounded to 0–10053.0670%Provisional rank
Grok 4.6 · xhigh66.7420.78Estimated: Estimated from grok-4.6 · high: 20.30 × (66.74 / 65.19), bounded to 0–10066.74Estimated: DeepSWE-only proxy: no same-model measured anchor available66.74Estimated: DeepSWE-only proxy: no same-model measured anchor available66.74Estimated: DeepSWE-only proxy: no same-model measured anchor available52.9540%Provisional rank
Grok 4.6 · high65.1920.3065.19Estimated: DeepSWE-only proxy: no same-model measured anchor available65.19Estimated: DeepSWE-only proxy: no same-model measured anchor available65.19Estimated: DeepSWE-only proxy: no same-model measured anchor available51.7270%Provisional rank
Sonnet 5 · max53.8512.4287.47Estimated: Estimated from claude-sonnet-5 · xhigh: 80.68 × (53.85 / 49.67), bounded to 0–10096.15Estimated: Estimated from claude-sonnet-5 · xhigh: 88.69 × (53.85 / 49.67), bounded to 0–10069.23Estimated: Estimated from claude-sonnet-5 · xhigh: 63.86 × (53.85 / 49.67), bounded to 0–10051.4670%Provisional rank
Opus 4.8 · xhigh54.3621.79Estimated: Estimated from claude-opus-4-8 · max: 23.64 × (54.36 / 58.97), bounded to 0–10075.43Estimated: Estimated from claude-opus-4-8 · max: 81.83 × (54.36 / 58.97), bounded to 0–10082.22Estimated: Estimated from claude-opus-4-8 · max: 89.19 × (54.36 / 58.97), bounded to 0–10066.40Estimated: Estimated from claude-opus-4-8 · max: 72.03 × (54.36 / 58.97), bounded to 0–10051.1440%Provisional rank
GPT-5.6 Terra · xhigh60.1818.60Estimated: Estimated from gpt-5.6-terra · max: 21.52 × (60.18 / 69.62), bounded to 0–10067.63Estimated: Estimated from gpt-5.6-terra · max: 78.25 × (60.18 / 69.62), bounded to 0–10078.34Estimated: Estimated from gpt-5.6-terra · max: 90.63 × (60.18 / 69.62), bounded to 0–10055.85Estimated: Estimated from gpt-5.6-terra · max: 64.62 × (60.18 / 69.62), bounded to 0–10050.4240%Provisional rank
Gemini 3.5 Flash · highLiveBench: gemini-3.5-flash-high36.0636.06Estimated: DeepSWE-only proxy: no same-model measured anchor available78.1882.0075.6048.9570%Provisional rank
GPT-5.6 Luna · xhigh56.8614.61Estimated: Estimated from gpt-5.6-luna · max: 17.27 × (56.86 / 67.19), bounded to 0–10070.17Estimated: Estimated from gpt-5.6-luna · max: 82.91 × (56.86 / 67.19), bounded to 0–10072.48Estimated: Estimated from gpt-5.6-luna · max: 85.64 × (56.86 / 67.19), bounded to 0–10050.88Estimated: Estimated from gpt-5.6-luna · max: 60.12 × (56.86 / 67.19), bounded to 0–10047.4440%Provisional rank
Opus 4.8 · high51.7720.75Estimated: Estimated from claude-opus-4-8 · max: 23.64 × (51.77 / 58.97), bounded to 0–10071.83Estimated: Estimated from claude-opus-4-8 · max: 81.83 × (51.77 / 58.97), bounded to 0–10078.30Estimated: Estimated from claude-opus-4-8 · max: 89.19 × (51.77 / 58.97), bounded to 0–10063.23Estimated: Estimated from claude-opus-4-8 · max: 72.03 × (51.77 / 58.97), bounded to 0–10048.7040%Provisional rank
Sonnet 5 · xhighLiveBench: claude-sonnet-5-xhigh-effort49.6711.46Estimated: Estimated from claude-sonnet-5 · max: 12.42 × (49.67 / 53.85), bounded to 0–10080.6888.6963.8647.4770%Provisional rank
Sonnet 5 · high48.2311.12Estimated: Estimated from claude-sonnet-5 · max: 12.42 × (48.23 / 53.85), bounded to 0–10078.35Estimated: Estimated from claude-sonnet-5 · xhigh: 80.68 × (48.23 / 49.67), bounded to 0–10086.13Estimated: Estimated from claude-sonnet-5 · xhigh: 88.69 × (48.23 / 49.67), bounded to 0–10062.01Estimated: Estimated from claude-sonnet-5 · xhigh: 63.86 × (48.23 / 49.67), bounded to 0–10046.0940%Provisional rank
Opus 4.8 · medium48.6719.51Estimated: Estimated from claude-opus-4-8 · max: 23.64 × (48.67 / 58.97), bounded to 0–10067.53Estimated: Estimated from claude-opus-4-8 · max: 81.83 × (48.67 / 58.97), bounded to 0–10073.61Estimated: Estimated from claude-opus-4-8 · max: 89.19 × (48.67 / 58.97), bounded to 0–10059.45Estimated: Estimated from claude-opus-4-8 · max: 72.03 × (48.67 / 58.97), bounded to 0–10045.7940%Provisional rank
GPT-5.6 Terra · high53.7616.62Estimated: Estimated from gpt-5.6-terra · max: 21.52 × (53.76 / 69.62), bounded to 0–10060.42Estimated: Estimated from gpt-5.6-terra · max: 78.25 × (53.76 / 69.62), bounded to 0–10069.99Estimated: Estimated from gpt-5.6-terra · max: 90.63 × (53.76 / 69.62), bounded to 0–10049.90Estimated: Estimated from gpt-5.6-terra · max: 64.62 × (53.76 / 69.62), bounded to 0–10045.0540%Provisional rank
Gemini 3.7 Flash · low53.769.23Estimated: Estimated from gemini-3.7-flash · high: 11.21 × (53.76 / 65.27), bounded to 0–10064.98Estimated: Estimated from gemini-3.7-flash · high: 78.89 × (53.76 / 65.27), bounded to 0–10072.32Estimated: Estimated from gemini-3.7-flash · high: 87.80 × (53.76 / 65.27), bounded to 0–10065.84Estimated: Estimated from gemini-3.7-flash · high: 79.93 × (53.76 / 65.27), bounded to 0–10044.5540%Provisional rank
GLM 5.2 · max43.7843.78Estimated: DeepSWE-only proxy: no same-model measured anchor available43.78Estimated: DeepSWE-only proxy: no same-model measured anchor available43.78Estimated: DeepSWE-only proxy: no same-model measured anchor available43.78Estimated: DeepSWE-only proxy: no same-model measured anchor available43.7840%Provisional rank
Kimi K2.7 CodeLiveBench: kimi-k2.7-code30.5330.53Estimated: DeepSWE-only proxy: no same-model measured anchor available73.9682.8156.2943.5670%Provisional rank
Grok 4.5 · high53.7612.4253.76Estimated: DeepSWE-only proxy: no same-model measured anchor available53.76Estimated: DeepSWE-only proxy: no same-model measured anchor available53.76Estimated: DeepSWE-only proxy: no same-model measured anchor available41.3670%Provisional rank
GPT-5.6 Sol · low45.3523.26Estimated: Estimated from gpt-5.6-sol · max: 37.27 × (45.35 / 72.67), bounded to 0–10052.39Estimated: Estimated from gpt-5.6-sol · max: 83.94 × (45.35 / 72.67), bounded to 0–10057.20Estimated: Estimated from gpt-5.6-sol · max: 91.65 × (45.35 / 72.67), bounded to 0–10044.84Estimated: Estimated from gpt-5.6-sol · max: 71.85 × (45.35 / 72.67), bounded to 0–10040.9440%Provisional rank
GPT-5.6 Luna · high44.2511.37Estimated: Estimated from gpt-5.6-luna · max: 17.27 × (44.25 / 67.19), bounded to 0–10054.61Estimated: Estimated from gpt-5.6-luna · max: 82.91 × (44.25 / 67.19), bounded to 0–10056.40Estimated: Estimated from gpt-5.6-luna · max: 85.64 × (44.25 / 67.19), bounded to 0–10039.59Estimated: Estimated from gpt-5.6-luna · max: 60.12 × (44.25 / 67.19), bounded to 0–10036.9240%Provisional rank
Opus 4.8 · low40.8016.35Estimated: Estimated from claude-opus-4-8 · max: 23.64 × (40.80 / 58.97), bounded to 0–10056.61Estimated: Estimated from claude-opus-4-8 · max: 81.83 × (40.80 / 58.97), bounded to 0–10061.70Estimated: Estimated from claude-opus-4-8 · max: 89.19 × (40.80 / 58.97), bounded to 0–10049.83Estimated: Estimated from claude-opus-4-8 · max: 72.03 × (40.80 / 58.97), bounded to 0–10038.3840%Provisional rank
Sonnet 5 · medium39.789.18Estimated: Estimated from claude-sonnet-5 · max: 12.42 × (39.78 / 53.85), bounded to 0–10064.62Estimated: Estimated from claude-sonnet-5 · xhigh: 80.68 × (39.78 / 49.67), bounded to 0–10071.03Estimated: Estimated from claude-sonnet-5 · xhigh: 88.69 × (39.78 / 49.67), bounded to 0–10051.14Estimated: Estimated from claude-sonnet-5 · xhigh: 63.86 × (39.78 / 49.67), bounded to 0–10038.0240%Provisional rank
GLM 5.2 · high36.2836.28Estimated: DeepSWE-only proxy: no same-model measured anchor available36.28Estimated: DeepSWE-only proxy: no same-model measured anchor available36.28Estimated: DeepSWE-only proxy: no same-model measured anchor available36.28Estimated: DeepSWE-only proxy: no same-model measured anchor available36.2840%Provisional rank
GPT-5.6 Luna · low1.550.40Estimated: Estimated from gpt-5.6-luna · max: 17.27 × (1.55 / 67.19), bounded to 0–1001.91Estimated: Estimated from gpt-5.6-luna · max: 82.91 × (1.55 / 67.19), bounded to 0–1001.97Estimated: Estimated from gpt-5.6-luna · max: 85.64 × (1.55 / 67.19), bounded to 0–1001.39Estimated: Estimated from gpt-5.6-luna · max: 60.12 × (1.55 / 67.19), bounded to 0–1001.2940%Provisional rank
Grok 4.6 · low41.6512.97Estimated: Estimated from grok-4.6 · high: 20.30 × (41.65 / 65.19), bounded to 0–10041.65Estimated: DeepSWE-only proxy: no same-model measured anchor available41.65Estimated: DeepSWE-only proxy: no same-model measured anchor available41.65Estimated: DeepSWE-only proxy: no same-model measured anchor available33.0440%Provisional rank
Gemini 3.1 Pro Preview · highLiveBench: gemini-3.1-pro-preview-high11.7311.73Estimated: DeepSWE-only proxy: no same-model measured anchor available76.4584.0079.1032.0370%Provisional rank
GPT-5.6 Terra · medium35.1110.85Estimated: Estimated from gpt-5.6-terra · max: 21.52 × (35.11 / 69.62), bounded to 0–10039.46Estimated: Estimated from gpt-5.6-terra · max: 78.25 × (35.11 / 69.62), bounded to 0–10045.71Estimated: Estimated from gpt-5.6-terra · max: 90.63 × (35.11 / 69.62), bounded to 0–10032.59Estimated: Estimated from gpt-5.6-terra · max: 64.62 × (35.11 / 69.62), bounded to 0–10029.4240%Provisional rank
Sonnet 4.6 · high29.9329.93Estimated: DeepSWE-only proxy: no same-model measured anchor available29.93Estimated: DeepSWE-only proxy: no same-model measured anchor available29.93Estimated: DeepSWE-only proxy: no same-model measured anchor available29.93Estimated: DeepSWE-only proxy: no same-model measured anchor available29.9340%Provisional rank
Sonnet 5 · low30.517.04Estimated: Estimated from claude-sonnet-5 · max: 12.42 × (30.51 / 53.85), bounded to 0–10049.56Estimated: Estimated from claude-sonnet-5 · xhigh: 80.68 × (30.51 / 49.67), bounded to 0–10054.49Estimated: Estimated from claude-sonnet-5 · xhigh: 88.69 × (30.51 / 49.67), bounded to 0–10039.23Estimated: Estimated from claude-sonnet-5 · xhigh: 63.86 × (30.51 / 49.67), bounded to 0–10029.1640%Provisional rank
GPT-5.5 · low26.9926.99Estimated: DeepSWE-only proxy: no same-model measured anchor available33.08Estimated: Estimated from gpt-5.5 · xhigh: 82.15 × (26.99 / 67.04), bounded to 0–10036.10Estimated: Estimated from gpt-5.5 · xhigh: 89.65 × (26.99 / 67.04), bounded to 0–10028.48Estimated: Estimated from gpt-5.5 · xhigh: 70.73 × (26.99 / 67.04), bounded to 0–10028.8940%Provisional rank
GPT-5.6 Terra · low24.057.43Estimated: Estimated from gpt-5.6-terra · max: 21.52 × (24.05 / 69.62), bounded to 0–10027.03Estimated: Estimated from gpt-5.6-terra · max: 78.25 × (24.05 / 69.62), bounded to 0–10031.31Estimated: Estimated from gpt-5.6-terra · max: 90.63 × (24.05 / 69.62), bounded to 0–10022.32Estimated: Estimated from gpt-5.6-terra · max: 64.62 × (24.05 / 69.62), bounded to 0–10020.1540%Provisional rank
GPT-5.6 Luna · medium11.282.90Estimated: Estimated from gpt-5.6-luna · max: 17.27 × (11.28 / 67.19), bounded to 0–10013.92Estimated: Estimated from gpt-5.6-luna · max: 82.91 × (11.28 / 67.19), bounded to 0–10014.38Estimated: Estimated from gpt-5.6-luna · max: 85.64 × (11.28 / 67.19), bounded to 0–10010.10Estimated: Estimated from gpt-5.6-luna · max: 60.12 × (11.28 / 67.19), bounded to 0–1009.4240%Provisional rank