Claude Opus 5.5 vs Gemini 4 Argon
Choose two or three models. Copy the URL to share your selection.
| Comparison | Claude Opus 5.5claude-opus-5-5 | Gemini 4 Argongemini-4-argon |
|---|---|---|
| Type | Chat | Chat |
| Best for | Complex reasoning, multi-step analysis and difficult coding | Long-horizon coding, agents and million-token documents |
| TokenRex price | $4.00 in / $20.00 out per 1M tokens | $2.00 in / $10.00 out per 1M tokens |
| Cached input / 1M tokens | $0.20 per 1M tokens | $0.10 per 1M tokens |
| Cache write / 1M tokens | $5.00 per 1M tokens | No extra fee |
| Cached output | Not offered · standard output rate applies | Not offered · standard output rate applies |
| Market price | $4.00 in / $20.00 out per 1M tokens | $2.00 in / $10.00 out per 1M tokens |
| Context window (provider) | 1,000,000 tokens | 1,048,576 tokens |
| Input token limit (provider) | Within the context window | Within the context window |
| Maximum output | 128,000 tokens | 65,536 tokens |
| Knowledge cutoff | Jun 2026 | March 2026 |
| Vendor input modalities | Text, Image | Text, Image, Audio, Video |
| Vendor output modalities | Text | Text |
| TokenRex endpoints | POST /v1/chat/completions · POST /v1/messages | POST /v1/chat/completions |
| TokenRex input | Text, Image/PDF content through the chat bridge | — |
| TokenRex output | Text | — |
| Benchmarks | ||
| Benchmark evidence | 17 source-linked results | 19 source-linked results |
| FrontierCode v1.1 (Main) | 54.4% Provider launch evaluation with adaptive thinking at max effort and production safeguards; not the default medium catalog configuration. Anthropic · Claude Opus 5.5 launch evaluation54.64% Cognition FrontierCode 1.1; Claude Opus 5.5; medium effort; claude-code harness; Main subset (100 tasks). Published weighted rubric score (new_score × 100), not the all-or-nothing pass rate; solution-source violations score zero. Source field new_score=0.5464. Publisher evaluation, not a TokenRex measurement; provider effort configuration is not a guarantee of the API's fixed thinking budget. Cognition · FrontierCode 1.1 original leaderboard | — |
| CursorBench 4.0 | 57.8% Provider launch evaluation with adaptive thinking at max effort and production safeguards; not the default medium catalog configuration. Anthropic · Claude Opus 5.5 launch evaluation56% Row explicitly labeled 'Opus 5.5 Extra High' (xhigh) at 56.0%, $6.98/task. Kept distinct from the existing 57.8% entry, which is Anthropic's own self-reported max-effort figure from the launch page, and from CursorBench's own 'Opus 5.5 Max' row (57.8%, $13.43/task) — same number as Anthropic's but a separately sourced/priced data point. Cursor · CursorBench 4.0 public leaderboard | — |
| AutomationBench | 40% Provider launch evaluation with adaptive thinking at max effort and production safeguards; not the default medium catalog configuration. Anthropic · Claude Opus 5.5 launch evaluation | 51.3% Self-reported by Google in the Gemini 4 Argon model card (Score), October 2026. Google DeepMind · Gemini 4 Argon model card |
| Humanity’s Last Exam (with tools) | 67.7% Provider launch evaluation with adaptive thinking at max effort and production safeguards; not the default medium catalog configuration. Anthropic · Claude Opus 5.5 launch evaluation | — |
| Terminal-Bench-Science 0.1 | 58.7% Provider launch evaluation with adaptive thinking at max effort and production safeguards; not the default medium catalog configuration. Anthropic · Claude Opus 5.5 launch evaluation62% Explicitly the xhigh effort configuration ('Claude Opus 5.5 (xhigh)'); AA also separately reports 59% at max effort for the same model on this benchmark — kept distinct and not merged. Independent AA harness, not Anthropic's own 58.7% (max effort) figure already on file. Artificial Analysis · Terminal-Bench-Science 0.1 leaderboard launch post (LinkedIn) | — |
| OSWorld 2.1 (partial) | 81.8% Provider launch evaluation with adaptive thinking at max effort and production safeguards; not the default medium catalog configuration. Anthropic · Claude Opus 5.5 launch evaluation | — |
| Chartography (with tools) | 89% Provider launch evaluation with adaptive thinking at max effort and production safeguards; not the default medium catalog configuration. Anthropic · Claude Opus 5.5 launch evaluation | — |
| Terminal-Bench 4.0 | 66.4% Launch report explicitly states xhigh effort, standard error ±2.6 points; production safeguards. Not a medium or fast-mode measurement. Anthropic · Claude Opus 5.5 launch evaluation60% Explicitly the Xhigh effort configuration ('Claude Opus 5.5 (Xhigh, Default Fallback)') on AA's independently-run Terminal-Bench 4.0 harness — matches the requested Opus 5.5 xhigh condition exactly, not max/fast/default. Distinct from Anthropic's own self-reported xhigh figure of 66.4% already on file (different methodology/harness) and from AA's narrative max-effort figure of 59.6% quoted in AA's launch article. Artificial Analysis · Opus 5.5 (Xhigh) vs Opus 5 (High) comparison | — |
| Artificial Analysis Intelligence Index | 58index Independent AA run, 'Claude Opus 5.5 (Max, Default Fallback)' configuration — AA's max-effort-with-fallback harness, not the catalog medium default. Not comparable to the Xhigh config (56) or High config (54) also reported by AA. Artificial Analysis · Claude Opus 5.5 release page | — |
| Humanity's Last Exam | 61.4% Independent AA evaluation, reported for the top-scoring (max-effort/fallback) AA configuration used for the Intelligence Index; AA states this is the new best score on this eval, ahead of Fable 5.1's prior 59.1%. Effort level for this specific sub-score is implied (Intelligence Index default config), not separately confirmed for every effort tier. Artificial Analysis · 'Claude Opus 5.5 takes the top spot' article | — |
| SciCode | 66.9% Independent AA evaluation under the Intelligence Index max-effort/fallback configuration; AA reports this surpasses Fable 5.1's prior best of 63.1%. Artificial Analysis · 'Claude Opus 5.5 takes the top spot' article | — |
| AA-Briefcase v1.1 (Elo) | 1822Elo AA's private frontier knowledge-work evaluation using the open-source Stirrup reference agent harness; max-effort/fallback configuration (the configuration AA headlines for Opus 5.5). Not the Xhigh-specific Briefcase score (1768) also published by AA. Artificial Analysis · 'Claude Opus 5.5 takes the top spot' article | — |
| FrontierCode v1.1 (Extended) | 65.27% Cognition FrontierCode 1.1; Claude Opus 5.5; medium effort; claude-code harness; Extended subset (150 tasks). Published weighted rubric score (new_score × 100), not the all-or-nothing pass rate; solution-source violations score zero. Source field new_score=0.6527. Publisher evaluation, not a TokenRex measurement; provider effort configuration is not a guarantee of the API's fixed thinking budget. Cognition · FrontierCode 1.1 original leaderboard | — |
| Vals Index | — | 68.9% Self-reported by Google in the Gemini 4 Argon model card, October 2026. Google DeepMind · Gemini 4 Argon model card |
| Vals Finance Agent v2 | — | 65.4% Self-reported by Google in the Gemini 4 Argon model card, October 2026. Google DeepMind · Gemini 4 Argon model card |
| Harvey's Legal Agent Benchmark | — | 19.6% Self-reported by Google in the Gemini 4 Argon model card, October 2026. Google DeepMind · Gemini 4 Argon model card |
| DeepSWE v1.1 | — | 77.9% Self-reported by Google in the Gemini 4 Argon model card, October 2026. Google DeepMind · Gemini 4 Argon model card |
| FrontierSWE v2 | — | 55% Self-reported by Google in the Gemini 4 Argon model card, October 2026. Google DeepMind · Gemini 4 Argon model card |
| Vibe Code Bench | — | 91.9% Self-reported by Google in the Gemini 4 Argon model card, October 2026. Google DeepMind · Gemini 4 Argon model card |
| Terminal-bench 4.0 | — | 57.4% Self-reported by Google in the Gemini 4 Argon model card, October 2026. Google DeepMind · Gemini 4 Argon model card |
| PostTrainBench | — | 45.3% Self-reported by Google in the Gemini 4 Argon model card, October 2026. Google DeepMind · Gemini 4 Argon model card |
| Terminal-Bench Science 0.1 | — | 57.6% Self-reported by Google in the Gemini 4 Argon model card, October 2026. Google DeepMind · Gemini 4 Argon model card |
| LABBench 2 | — | 88.8% Self-reported by Google in the Gemini 4 Argon model card, October 2026. Google DeepMind · Gemini 4 Argon model card |
| RiemannBench | — | 76% Self-reported by Google in the Gemini 4 Argon model card, October 2026. Google DeepMind · Gemini 4 Argon model card |
| GraphWalks BFS (up to 128K) | — | 99.7% Self-reported by Google in the Gemini 4 Argon model card (F1), October 2026. Google DeepMind · Gemini 4 Argon model card |
| GraphWalks BFS (256K to 1M) | — | 84.2% Self-reported by Google in the Gemini 4 Argon model card (F1), October 2026. Google DeepMind · Gemini 4 Argon model card |
| Agent's Last Exam | — | 39.5% Self-reported by Google in the Gemini 4 Argon model card (Pass rate), October 2026. Google DeepMind · Gemini 4 Argon model card |
| OSWorld-2.0 (offline subset) | — | 69.2% Self-reported by Google in the Gemini 4 Argon model card (Partial score), October 2026. Google DeepMind · Gemini 4 Argon model card |
| Chartography | — | 71.6% Self-reported by Google in the Gemini 4 Argon model card, October 2026. Google DeepMind · Gemini 4 Argon model card |
| LVBench | — | 91.7% Self-reported by Google in the Gemini 4 Argon model card, October 2026. Google DeepMind · Gemini 4 Argon model card |
| CWE-bench v1 | — | 68% Self-reported by Google in the Gemini 4 Argon model card, October 2026. Google DeepMind · Gemini 4 Argon model card |
No overall score, winner or cost-versus-score frontier is calculated without comparable, attributable benchmark evidence. Different test configurations appear separately. Vendor capabilities are distinct from TokenRex API support.
— means no published value or no matching capability. Provider token limits are not tested gateway guarantees; TokenRex media controls and response formats are listed separately.
Claude Opus 5.5
Sources and verification
Provider specifications describe the upstream model, not a guarantee of every feature through TokenRex. Prices come from the enabled catalog; benchmark results belong to the exact tested model.
- Provider model specifications
Vendor-reported
Last checked - Anthropic · Claude Opus 5.5 launch evaluation
Vendor-reported
Last checked - Artificial Analysis · Claude Opus 5.5 release page
Third-party report
Last checked - Artificial Analysis · 'Claude Opus 5.5 takes the top spot' article
Third-party report
Last checked - Artificial Analysis · Opus 5.5 (Xhigh) vs Opus 5 (High) comparison
Third-party report
Last checked - Artificial Analysis · Terminal-Bench-Science 0.1 leaderboard launch post (LinkedIn)
Third-party report
Last checked - Cursor · CursorBench 4.0 public leaderboard
Third-party report
Last checked - Cognition · FrontierCode 1.1 original leaderboard
Third-party report
Last checked
Research notes (1)
- Published provider limits; maximum-capacity requests have not been independently exercised through TokenRex. Thinking-level variants use the same provider model; reasoning and answer tokens share the output budget.
Gemini 4 Argon
Sources and verification
Provider specifications describe the upstream model, not a guarantee of every feature through TokenRex. Prices come from the enabled catalog; benchmark results belong to the exact tested model.
- Gemini API · Gemini 4 Argon model specifications
Vendor-reported
Last checked - Google DeepMind · Gemini 4 Argon model card
Vendor-reported
Last checked - Google Blog · Introducing Gemini 4 Argon
Vendor-reported
Last checked
Research notes (2)
- Published provider limits; maximum-capacity requests have not been independently exercised through TokenRex. Standard requests allow up to 65,536 output tokens. Google's long-decode mode (output up to 1,048,576 tokens minus input) is enabled with background=true — the TokenRex chat endpoint accepts this parameter (verified live on 2026-10-14), but a full 1M-token output has not been exercised, so the default 65,536-token output limit is shown.
- All benchmark scores are self-reported by Google in the official model card; TokenRex has not re-run them.
Popular comparisons
- GPT-6 Sol vs Claude Opus 5.5
- GPT-6 Sol vs Gemini 4 Argon
- Claude Opus 5.5 vs Gemini 4 Argon
- GPT-6.1 Sol vs GPT-6 Sol
- GPT-6 Astra vs Claude Sonnet 5
- GPT-6 Astra vs Gemini 3.1 Pro
- Claude Sonnet 5 vs Gemini 3.1 Pro
- Claude Opus 5.5 vs Claude Sonnet 5
- GPT-5.5 vs Claude Sonnet 5
- GPT-5.5 vs Gemini 2.5 Pro
- GPT-5 vs Gemini 2.5 Pro
- GPT-6 Luna vs Gemini 3.8 Flash
- Claude Haiku 4.5 vs Gemini 3.8 Flash
- GPT-5 Mini vs Gemini 2.5 Flash
- GPT-5 Nano vs Gemini 2.5 Flash Lite
- GPT-5.4 Mini vs Claude Haiku 4.5