Google DeepMind model card, results as of February 2026. · source
Single attempt; the card gives no Claude figure for this benchmark. · source
| Item | Value |
|---|---|
| Inputs | Text, images, audio, video; context window up to 1M tokens |
| Output | Text, 64K tokens |
| Based on | Gemini 3 Pro |
| Humanity's Last Exam (no tools) | 44.4% (Gemini 3 Pro 37.5%) |
| GPQA Diamond (no tools) | 94.3% |
| SWE-Bench Verified (single attempt) | 80.6% |
| Terminal-Bench 2.0 (Terminus-2) | 68.5% |
| tau2-bench Telecom | 99.3% |
| BrowseComp | 85.9% |
| MCP Atlas | 69.2% |
| MRCR v2 8-needle, 1M pointwise | 26.3% |
Google's own figures, results as of February 2026. The card lists no API price. · source
Partial outage now. Google Gemini is currently experiencing a partial outage
Unofficial, from StatusGator. Uptime is the share of the window not marked down (559 min down, 3277 min degraded); it covers the whole service, not this version alone.
Estimate from the version's published specs, benchmarks and reviews, 0-5; scores are only comparable within one criteria year. Scored as launched (19 Feb 2026, preview). Reasoning: Artificial Analysis Intelligence Index 57 at launch (v4.0; four points ahead of Claude Opus 4.6; The Decoder 21 Feb 2026). AA's pages now show lower numbers (30, 46) after later index versions, not used. Coding null: no Terminal-Bench 4.0; Google published Terminal-Bench 2.0 68.5% (Terminus-2) and SWE-Bench Verified 80.6%. Context 1M input, 64k output (DeepMind model card). Price $2 in / $12 out per 1M up to 200k = $4.50 blended (AA, Google). Speed null: no launch-time speed found (AA now lists 132 tokens/s, undated). Input text, image, audio, video.[1]
The model ranks first in six of ten categories, including agent-based coding, knowledge, scientific reasoning, and physics.
While the leap in logical reasoning and coding benchmarks is impressive, the 'soul' of the model — its emotional depth, empathy, creative flexibility, and nuance — seems to have been significantly reduced.
It is not necessarily faster, and in some modes, it is intentionally slower, taking a moment to chew on a problem before spitting out an answer.
Winner: Gemini wins for providing phrasing I'd actually use because it feels the most natural and immediately usable.
Photos from readers, kept apart from the pin's sourced media. Rate what you ate or saw, and on a restaurant pin tag the dishes right on your photo.
The first entry is always the pin's source. Overall confidence is a weighted average of how firmly each reference supports the start and end times used above; a reference counts half as much for every 180 days older than the newest.
Google's post "Gemini 3.1 Pro: A smarter model for your most complex tasks" is dated 19 February 2026 and says it is rolling out that day to developers, enterprises and consumers.
The Wikipedia article lists Gemini 3.1 Pro as released in preview on 19 February 2026.
DeepMind's model card gives the benchmark table against the model's predecessor and named rivals.
Something missing or wrong? Say it in your own words: a link that backs this pin up, a different start or end date and why, or a fact it lacks or gets wrong. The AI checks it against this pin's sources, searches for better ones, and adds any page that backs you up. The pin's own sources still count most. A picture that shows something else, or shows it badly, is looked at too, and moved down or replaced.
| Benchmark | Gemini 3.1 Pro | Gemini 3 Pro | Sonnet 4.6 | Opus 4.6 | GPT-5.2 | GPT-5.3-Codex |
|---|---|---|---|---|---|---|
| Humanity's Last Exam (Academic reasoning (full set, text + MM), No tools) | 44.4% | 37.5% | 33.2% | 40.0% | 34.5% | — |
| Humanity's Last Exam (Academic reasoning (full set, text + MM), Search (blocklist) + Code) | 51.4% | 45.8% | 49.0% | 53.1% | 45.5% | — |
| ARC-AGI-2 (Abstract reasoning puzzles, ARC Prize Verified) | 77.1% | 31.1% | 58.3% | 68.8% | 52.9% | — |
| GPQA Diamond (Scientific knowledge, No tools) | 94.3% | 91.9% | 89.9% | 91.3% | 92.4% | — |
| Terminal-Bench 2.0 (Agentic terminal coding, Terminus-2 harness) | 68.5% | 56.9% | 59.1% | 65.4% | 54.0% | 64.7% |
| Terminal-Bench 2.0 (Agentic terminal coding, Other best self-reported harness) | — | — | — | — | 62.2% | 77.3% |
| SWE-Bench Verified (Agentic coding, Single attempt) | 80.6% | 76.2% | 79.6% | 80.8% | 80.0% | — |
| SWE-Bench Pro (Public) (Diverse agentic coding tasks, Single attempt) | 54.2% | 43.3% | — | — | 55.6% | 56.8% |
| LiveCodeBench Pro (Competitive coding problems from Codeforces, ICPC, and IOI, Elo) | 2887 | 2439 | — | — | 2393 | — |
| SciCode (Scientific research coding) | 59% | 56% | 47% | 52% | 52% | — |
| APEX-Agents (Long horizon professional tasks) | 33.5% | 18.4% | — | 29.8% | 23.0% | — |
| GDPval-AA Elo (Expert tasks) | 1317 | 1195 | 1633 | 1606 | 1462 | — |
| τ2-bench (Agentic and tool use, Retail) | 90.8% | 85.3% | 91.7% | 91.9% | 82.0% | — |
| τ2-bench (Agentic and tool use, Telecom) | 99.3% | 98.0% | 97.9% | 99.3% | 98.7% | — |
| MCP Atlas (Multi-step workflows using MCP) | 69.2% | 54.1% | 61.3% | 59.5% | 60.6% | — |
| BrowseComp (Agentic search, Search + Python + Browse) | 85.9% | 59.2% | 74.7% | 84.0% | 65.8% | — |
| MMMU-Pro (Multimodal understanding and reasoning, No tools) | 80.5% | 81.0% | 74.5% | 73.9% | 79.5% | — |
| MMMLU (Multilingual Q&A) | 92.6% | 91.8% | 89.3% | 91.1% | 89.6% | — |
| MRCR v2 (8-needle) (Long context performance, 128k (average)) | 84.9% | 77.0% | 84.9% | 84.0% | 83.8% | — |
| MRCR v2 (8-needle) (Long context performance, 1M (pointwise)) | 26.3% | 26.3% | — | — | — | — |