ModelComp

Compare AI coding models on what actually matters

Tool use, reasoning, context, multimodal, coding and cost — normalized to 1–100 from public benchmarks so current and future models can be compared side by side.

Compare modelsHow scoring works

Compare models

Pick up to three models. The hexagon shows the six scored dimensions (0–100); Overall Score is listed in the legend and table. Decision models (e.g. Jev) can't be picked here — their scores aren't comparable; see the decision shelf below.

Radar chart comparing Muse Spark 1.3, Claude Opus 5.5, Gemini 3.8 FlashHexagonal radar with axes Tool use, Reasoning, Context window, Multimodal, Coding and Cost efficiency, scaled 0 to 100 with visual emphasis on the top end (the outer half of each axis covers 75 to 100). Exact values are listed in the data table below the chart.255075100Muse Spark 1.3 (Overall 94)Claude Opus 5.5 (Overall 93)Gemini 3.8 Flash (Overall 92.5)Muse Spark 1.3 — Tool use: 95.3 (Free OpenCode Zen tier (Contributor) in exchange for training-data consent — same weights as paid Standard; Max reasoning effort is Standard-tier only)Muse Spark 1.3 — Reasoning: 93.4 (Free OpenCode Zen tier (Contributor) in exchange for training-data consent — same weights as paid Standard; Max reasoning effort is Standard-tier only)Muse Spark 1.3 — Context: 1,048,576 (1M) total; 131,072 max output (Free OpenCode Zen tier (Contributor) in exchange for training-data consent — same weights as paid Standard; Max reasoning effort is Standard-tier only)Muse Spark 1.3 — Cost: Free OpenCode Zen tier; Contributor $0.10/$0.20; Standard & Max effort $1.25/$4.25 per 1M (Free OpenCode Zen tier (Contributor) in exchange for training-data consent — same weights as paid Standard; Max reasoning effort is Standard-tier only)Muse Spark 1.3 — Coding: 95.2 (Free OpenCode Zen tier (Contributor) in exchange for training-data consent — same weights as paid Standard; Max reasoning effort is Standard-tier only)Muse Spark 1.3 — Multimodal: Text, image, video, PDF in; text out (Free OpenCode Zen tier (Contributor) in exchange for training-data consent — same weights as paid Standard; Max reasoning effort is Standard-tier only)Claude Opus 5.5 — Tool use: 94.6Claude Opus 5.5 — Reasoning: 94.7Claude Opus 5.5 — Context: 1M / 128K outClaude Opus 5.5 — Cost: Paid-tier pricingClaude Opus 5.5 — Coding: 95.4Claude Opus 5.5 — Multimodal: Text, image in; text outGemini 3.8 Flash — Tool use: 90.4 (Free tier available on Google AI Studio and OpenCode Zen with standard rate limits)Gemini 3.8 Flash — Reasoning: 91.4 (Free tier available on Google AI Studio and OpenCode Zen with standard rate limits)Gemini 3.8 Flash — Context: 1,048,576 (1M) (Free tier available on Google AI Studio and OpenCode Zen with standard rate limits)Gemini 3.8 Flash — Cost: Free tier available; Paid-tier pricing (Free tier available on Google AI Studio and OpenCode Zen with standard rate limits)Gemini 3.8 Flash — Coding: 90.7 (Free tier available on Google AI Studio and OpenCode Zen with standard rate limits)Gemini 3.8 Flash — Multimodal: Text, image, audio, PDF in; text out (Free tier available on Google AI Studio and OpenCode Zen with standard rate limits)Performance on agent benchmarks like Terminal-Bench, Tau3, and Claw-Eval.ToolAbility to solve complex, multi-step problems, based on reasoning benchmarks. Higher reasoning results score higher.ReasonHow much input the model can process at once. Larger context windows score higher.ContextRelative input and output pricing. Lower-cost models score higher.CostPerformance on software engineering and coding benchmarks. Stronger coding ability scores higher.CodeSupport for non-text input and output, including images, audio, and video.Multi

Overall score — average of qualifying reports (raters above 84.9 Overall).

Muse Spark 1.3

Overall 94Free
Tool use
95.3
Reasoning
93.4
Context window
99.8
Cost efficiency
93.8
Coding
95.2
Multimodal
86.3
Overall Score
94
Context window
1,048,576 (1M) total; 131,072 max output
Pricing / 1M
  • Free OpenCode Zen tier
  • Contributor $0.10/$0.20
  • Standard & Max effort $1.25/$4.25

Claude Opus 5.5

Overall 93Paid
Tool use
94.6
Reasoning
94.7
Context window
96.9
Cost efficiency
50.6
Coding
95.4
Multimodal
83
Overall Score
93
Context window
1M / 128K out
Pricing / 1M
  • Paid-tier pricing

Gemini 3.8 Flash

Overall 92.5Free
Tool use
90.4
Reasoning
91.4
Context window
96.5
Cost efficiency
92.8
Coding
90.7
Multimodal
93
Overall Score
92.5
Context window
1,048,576 (1M)
Pricing / 1M
  • Free tier available; Paid-tier pricing

How scoring works

Scores are normalized interpretations from 1 to 100 (higher is better), based on public benchmarks. Overall Score is the rounded mean of the five quality dimensions — Cost efficiency is scored separately and never counts toward Overall.

  • Independently researched: each model is checked across multiple public sources, and every benchmark number cites its source. Missing data is marked as no verified score — never invented.
  • Averaged, not single-opinion: a model's scores are the mean across independent reporting agents; reports without verified benchmarks are quarantined and never counted.
  • Re-researched weekly: every report is checked against fresh benchmarks once it is older than seven days, and scores are refreshed whenever new verified evidence would change them.
  • Decision models are a separate paradigm: System-One models (e.g. Jev) return typed judgments instead of generated text, so their scores are never comparable to generative Overall. They render on their own shelf and can't be picked in model comparison.
  • Tool use: Terminal-Bench, Tau-bench, GDPval and tool-call efficiency.
  • Reasoning: GPQA, Humanity's Last Exam, long-context retrieval and intelligence indexes.
  • Context window: tiered by size; 1M+ scores 95–100, 200K scores around 70.
  • Cost efficiency: inverse pricing on the evaluated tier; $0 free tiers score 100.
  • Coding: SWE-bench, DeepSWE, LiveCodeBench, SciCode and terminal coding tasks.
  • Multimodal: text-only scores low; image, video, audio or PDF input scores higher.

All models

Compare AI models across six dimensions — tool use, reasoning, context, coding, multimodal, and cost — ranked by Overall score. Switch the results source to see the average or any individual reviewer's ratings, search to find a model, or open its page for the full breakdown. Decision models (typed-judgment systems like Jev) live on a separate shelf below — they answer with judgments instead of text, so their scores aren't comparable.

MetaMuse Spark 1.3

94

Free

Meta's frontier coding and long-horizon agentic model, 1M context / 131K output. Contributor/Free and Max are the same weights — only price and training-data consent differ.

Context:
1,048,576 (1M) total; 131,072 max output
Modalities:
Text, image, video, PDF in; text out
Pricing:
Free OpenCode Zen tier; Contributor $0.10/$0.20; Standard & Max effort $1.25/$4.25 per 1M
Verified:
2026-10-12

Tool 95.3 · Reasoning 93.4 · Context 99.8 · Cost 93.8 · Coding 95.2 · Multimodal 86.3

AnthropicClaude Opus 5.5

93

Paid fallback — no Free ID

Anthropic's Opus 5.5 (1M/128K, $4/$20) for long-horizon coding; TB4 66.4%, AA Index 58.

Context:
1M / 128K out
Modalities:
Text, image in; text out
Pricing:
Paid-tier pricing
Verified:
2026-10-13

Tool 94.6 · Reasoning 94.7 · Context 96.9 · Cost 50.6 · Coding 95.4 · Multimodal 83

GoogleGemini 3.8 Flash

92.5

Free

Google's most-intelligent Flash (1M, $0.75/$3.75 intro) for agentic coding/enterprise; Terminal-Bench 90.8% and GPQA 95.3%.

Context:
1,048,576 (1M)
Modalities:
Text, image, audio, PDF in; text out
Pricing:
Free tier available; Paid-tier pricing
Verified:
2026-10-13

Tool 90.4 · Reasoning 91.4 · Context 96.5 · Cost 92.8 · Coding 90.7 · Multimodal 93

GoogleGemini 4 Argon

91.9

Paid fallback — no Free ID

Google DeepMind's 1M-output frontier model for long-horizon coding and enterprise research; DeepSWE 77.9% and Vals Index 68.9% both #1, gated launch.

Context:
1M total (262K output; 1M via continuation)
Modalities:
Text, image, video in; text out
Pricing:
Intro $2/$10 per 1M; then $4/$20
Verified:
2026-10-13

Tool 89.8 · Reasoning 91.1 · Context 96.7 · Cost 71.5 · Coding 92.3 · Multimodal 89.5

GoogleGemini 3.1 Pro

91.7

Free

Google's Pro flagship (2M/64K) for research/multimodal agents; ARC-AGI-2 77.1% and GPQA 94.3%.

Context:
2M / 64K out
Modalities:
Text, image, audio, video, PDF in; text out
Pricing:
Free tier available; Paid-tier pricing
Verified:
2026-10-13

Tool 88.8 · Reasoning 92.8 · Context 95.8 · Cost 79 · Coding 87.5 · Multimodal 93.3

OpenAIGPT-6 Astra

91.5

Paid fallback — no Free ID

OpenAI's 1.05M-context Astra flagship for frontier reasoning and agents; FrontierMath 97.6%, GPQA 96%, gated Daybreak rollout as first Critical cyber tier.

Context:
1,050,000 (1M) / 128K out
Modalities:
Text, image in; text out
Pricing:
Paid $10/$50 per 1M (no Zen Free ID)
Verified:
2026-10-13

Tool 93.3 · Reasoning 95.6 · Context 97.1 · Cost 36.4 · Coding 93.1 · Multimodal 77.9

AnthropicClaude Opus 5

91.3

Paid fallback — no Free ID

Anthropic's Opus (1M/128K, $5/$25) for long autonomous coding; SWE-Pro 79.2%, succeeded by Opus 5.5.

Context:
1M / 128K out
Modalities:
Text, image, PDF in; text out
Pricing:
Paid $5/$25 per 1M (no Zen Free ID)
Verified:
2026-10-13

Tool 93.3 · Reasoning 92.7 · Context 95.1 · Cost 45.9 · Coding 94.2 · Multimodal 81.2

GoogleGemini 3.7 Flash

91.1

Free

Google's workhorse Flash (1M/64K, intro $0.75/$3.75) for coding/agents; Terminal-Bench 85.8% and HLE-Verified 53.6%.

Context:
1,048,576 (1M)
Modalities:
Text, image, audio, PDF in; text out
Pricing:
Free tier available; Paid-tier pricing
Verified:
2026-10-13

Tool 88.8 · Reasoning 89.5 · Context 95.8 · Cost 91.3 · Coding 88.7 · Multimodal 91.8

AnthropicClaude Fable 5.1

91

Paid fallback — no Free ID

Anthropic's Mythos-class model above Opus 5 (1M/128K, $10/$50) for long-horizon coding; SWE-Pro 81.2%.

Context:
1M / 128K out
Modalities:
Text, image, PDF in; text out
Pricing:
Paid $10/$50 per 1M (no Zen Free ID)
Verified:
2026-10-13

Tool 94.1 · Reasoning 94.6 · Context 96 · Cost 36.1 · Coding 93.7 · Multimodal 77.1

Decision models

System-One models return typed judgments (choice, score, yes/no probability) instead of generated text — for routing, triage, and guardrails inside agent pipelines. They are scored on different axes and are not comparable to generative Overall scores, so they are never listed in the table above or in the A/B/C compare slots.

Unknown vendorJJev 1.13

Decision

TypeSafe AI's System-One typed-decision model (not generative text) with 64K budget for routing/triage; 136.5ms median, $0.042/M in, output free.

State budget:
64K state budget (state + questions; 32K state + longest question)
I/O:
Text in → typed decision out (Choice/Score/Noul); no image, audio, or video
Pricing:
$0.042/M input; output free (no autoregressive generation)