DeepSeek V4 Pro 0813 vs Gemini 3.8 Flash vs Mistral Large 4
Gemini 3.8 Flash comes out ahead, 74 to 70 and 67 on our weighted score, though DeepSeek V4 Pro 0813 is 34% cheaper per token.Gemini 3.8 Flash comes out ahead, 86 to 77 and 73 on our weighted score, though DeepSeek V4 Pro 0813 is 34% cheaper per token.
DeepSeek
DeepSeek V4 Pro 0813
67/100- ECI156.0
- Price$0.66 / $1.98
- Context1M
- Our pick
Google
Gemini 3.8 Flash
74/100- ECI158.8
- Price$0.75 / $3.75
- Context1.05M
Mistral AI
Mistral Large 4
70/100- ECI154.6
- Price$0.70 / $2.10
- Context1.05M
DeepSeek
DeepSeek V4 Pro 0813
73/100- ECI156.0
- Price$0.66 / $1.98
- Context1M
- Our pick
Google
Gemini 3.8 Flash
86/100- ECI158.8
- Price$0.75 / $3.75
- Context1.05M
Mistral AI
Mistral Large 4
77/100- ECI154.6
- Price$0.70 / $2.10
- Context1.05M
Gemini 3.8 Flash is our pick
Gemini 3.8 Flash is the better all-round choice, scoring 74/100 against Mistral Large 4 (70) and DeepSeek V4 Pro 0813 (67). It leads on capability and inputs & features. The score weighs capability 50%, price 25%, inputs & features 15%, context window 10%.Gemini 3.8 Flash is the better all-round choice, scoring 86/100 against Mistral Large 4 (77) and DeepSeek V4 Pro 0813 (73). It leads on capability and inputs & features. The score weighs capability 67%, inputs & features 20%, context window 13%.
- CapabilityGemini 3.8 FlashCapabilities Index (ECI): Gemini 3.8 Flash 158.8 · DeepSeek V4 Pro 0813 156.0 · Mistral Large 4 154.6
- Lowest priceDeepSeek V4 Pro 0813DeepSeek V4 Pro 0813 $0.99 · Mistral Large 4 $1.05 · Gemini 3.8 Flash $1.50 per 1M tokens (3:1 blend)
- Longest contextGemini 3.8 Flash and Mistral Large 4Gemini 3.8 Flash 1,048,576 · Mistral Large 4 1,048,576 · DeepSeek V4 Pro 0813 1,000,000 tokens
- Widest inputsGemini 3.8 FlashDeepSeek V4 Pro 0813: Text · Gemini 3.8 Flash: Text, Images, PDFs, Audio, Video · Mistral Large 4: Text, Images
- Self-hostingDeepSeek V4 Pro 0813Publishes downloadable weights (MIT)
| Measure | Weight | DeepSeek V4 Pro 0813 | Gemini 3.8 Flash | Mistral Large 4 |
|---|---|---|---|---|
| CapabilityCapabilities Index (ECI) | 50% | 83 | 86 | 82 |
| Price | 25% | 48 | 40 | 47 |
| Inputs & features | 15% | 45 | 100 | 70 |
| Context window | 10% | 64 | 65 | 65 |
| Overall | 100% | 67/100 | 74/100 | 70/100 |
| Measure | Weight | DeepSeek V4 Pro 0813 | Gemini 3.8 Flash | Mistral Large 4 |
|---|---|---|---|---|
| CapabilityCapabilities Index (ECI) | 67% | 83 | 86 | 82 |
| Inputs & features | 20% | 45 | 100 | 70 |
| Context window | 13% | 64 | 65 | 65 |
| Overall | 100% | 73/100 | 86/100 | 77/100 |
Every spec in one table
Highlighted cells lead their row. Unavailable specifications are marked “Not published”. Publisher results include their evaluation settings and source.
| Specification | |||
|---|---|---|---|
| Capability | |||
| Capabilities Index (ECI) | 156.0 | 158.8 (best) | 154.6 |
| ECI rank | #43 of 159 | #25 of 159 (best) | #50 of 159 |
| ECI confidence interval | 154.1–158.1 | 156.8–161.0 | 151.8–157.4 |
| ECI evaluation date | Aug 13, 2026 | Sep 2, 2026 | Oct 29, 2026 |
| Documented capabilities | DeepSeek V4 Pro snapshot with million-token context and support for thinking and non-thinking modes | Google's most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows | Mistral's 1T-parameter multimodal MoE for reasoning, coding, and agentic work |
| Model parameters | Not published | Not published | Not published |
| Output modalities | Text | Text | Text |
| GPQA DiamondGraduate-level science questions | 91.7%max | 95.4%high (best) | 90.2%high |
| FrontierMath Tiers 1–3Research-level mathematics | 64.6%max | 68.4%high (best) | Not published |
| OTIS Mock AIME 2024–2025Competition mathematics | 98.6%max | 98.9%high | 99.4%high (best) |
| SimpleQA VerifiedShort factual questions | 52.9%max | 69.7%high (best) | 22.4%high |
| Publisher-reported benchmarks | |||
| Humanity's Last Examwithout tools | 42.7scoreReported result | Not published | Not published |
| Humanity's Last Examwith tools | 60scoreReported result | Not published | Not published |
| Terminal-Bench 2.1max effort | 87.9scoreReported result | Not published | Not published |
| NL2Repomax effort | 61.5scoreReported result | Not published | Not published |
| CyberGymmax effort | 83.3scoreReported result | Not published | Not published |
| DeepSWEmax effort | 62.7scoreReported result | Not published | Not published |
| Toolathlon-VerifiedPublisher-reported result | 74.1scoreReported result | Not published | Not published |
| Agents' Last ExamPublisher-reported result | 25.7scoreReported result | Not published | Not published |
| AutomationBench · publicPublisher-reported result | 31.8scoreReported result | Not published | Not published |
| DSBench-FullStack · internalPublisher-reported result | 71.1scoreReported result | Not published | Not published |
| DSBench-Hard · internalPublisher-reported result | 67.2scoreReported result | Not published | Not published |
| DeepSWE 1.1high thinking | Not published | 73.7pass@1Reported result | Not published |
| GDPval-AA 2Publisher-reported result | Not published | 1545EloReported result | Not published |
| Finance Agent 2Publisher-reported result | Not published | 61.4pass@1Reported result | Not published |
| Harvey Legal Agent BenchmarkPublisher-reported result | Not published | 10all-pass rateReported result | Not published |
| Terminal-Bench 2.1Publisher-reported result | Not published | 89.4pass@1Reported result | Not published |
| Terminal-Bench 4.0Publisher-reported result | Not published | 19.1pass@1Reported result | Not published |
| GDP.PDFPublisher-reported result | Not published | 35all-pass rateReported result | Not published |
| CharXiv Reasoningwithout tools | Not published | 86.2pass@1Reported result | Not published |
| LVBenchagentic | Not published | 87.8pass@1Reported result | Not published |
| LVBenchstatic; 1024 frames; without tools | Not published | 87.1pass@1Reported result | Not published |
| HLE-Verified · 1811 verified and revised itemsPublisher-reported result | Not published | 54.9accuracyReported result | Not published |
| OSWorld 2.0 · before 2026-08-08 patchbatched tool calls; best of 3 runs; 500 steps | Not published | 59partial scoreReported result | Not published |
| BioMysteryBench · human-solvablewith tools | Not published | 88.8pass@1Reported result | Not published |
| BioMysteryBench · human-difficultwith tools | Not published | 56.5pass@1Reported result | Not published |
| LAB-Bench 2 · 11 subtaskswith tools | Not published | 86.2macro-averageReported result | Not published |
| Price per million tokens | |||
| Input | $0.66 (best) | $0.75 | $0.70 |
| Output | $1.98 (best) | $3.75 | $2.10 |
| Cached input | $0.022 (best) | $0.075 | $0.07 |
| Blended (3:1) | $0.99 (best) | $1.50 | $1.05 |
| Long-context rate | Same rate | Same rate | Same rate |
| Price source | Official DeepSeek API | Official Google API | Official Mistral API |
| Limits | |||
| Context window | 1,000,000 tokens | 1,048,576 tokens (best) | 1,048,576 tokens (best) |
| Max input | Not published | Not published | Not published |
| Max output | 384,000 tokens (best) | 65,536 tokens | 262,144 tokens |
| Inputs and features | |||
| Text | Yes | Yes | Yes |
| Images | No | Yes | Yes |
| PDFs | No | Yes | No |
| Audio | No | Yes | No |
| Video | No | Yes | No |
| Reasoning | Yeslow · high · max | Yeslow · medium · high | Yeslow · medium · high |
| Tool calling | Yes | Yes | Yes |
| Structured output | Yes | Yes | Yes |
| File attachments | No | Yes | Yes |
| Temperature control | Yes | Yes | Yes |
| Reasoning control | effort | effort | effort |
| Availability | |||
| Weights | OpenMIT | Proprietary | Proprietary |
| API model ID | deepseek-v4-pro | gemini-3.8-flash | mistral-large-4 |
| API providers | 54 (best) | 42 | 16 |
| Released | Aug 13, 2026 | Sep 2, 2026 | Oct 6, 2026 |
| Knowledge cutoff | Not published | Not published | Not published |
| Publisher documentation | Publisher documentation · Model weights | Publisher documentation | Publisher documentation |
| Source metadata updated | Aug 22, 2026 | Sep 2, 2026 | Oct 6, 2026 |
| Data sources | models.dev · Epoch AI | models.dev · Epoch AI | models.dev · Epoch AI |
| Catalog checked | Nov 22, 2026 | Nov 22, 2026 | Nov 22, 2026 |
What would a month cost?
Enter your expected volume in millions of tokens. List prices only; caching and batch discounts would lower these.
DeepSeek V4 Pro 0813$10.56
Gemini 3.8 Flash$15.00
Mistral Large 4$11.20
Which should you choose?
Which is better: DeepSeek V4 Pro 0813, Gemini 3.8 Flash or Mistral Large 4?
Gemini 3.8 Flash is the better all-round choice, scoring 74/100 against Mistral Large 4 (70) and DeepSeek V4 Pro 0813 (67). It leads on capability and inputs & features. The score weighs capability 50%, price 25%, inputs & features 15%, context window 10%.Gemini 3.8 Flash is the better all-round choice, scoring 86/100 against Mistral Large 4 (77) and DeepSeek V4 Pro 0813 (73). It leads on capability and inputs & features. The score weighs capability 67%, inputs & features 20%, context window 13%.
Which is cheaper, DeepSeek V4 Pro 0813, Gemini 3.8 Flash or Mistral Large 4?
DeepSeek V4 Pro 0813 is cheaper at $0.66 input / $1.98 output per million tokens (official DeepSeek API price). Mistral Large 4 costs $0.70 input / $2.10 output per million tokens (official Mistral API price); Gemini 3.8 Flash costs $0.75 input / $3.75 output per million tokens (official Google API price). At a typical mix of three input tokens to one output token, that is $0.99 per million tokens for DeepSeek V4 Pro 0813 versus $1.05 for Mistral Large 4 (1.1× as much) and $1.50 for Gemini 3.8 Flash (1.5× as much).
Which scores higher on benchmarks?
Gemini 3.8 Flash scores higher on the Capabilities Index (ECI): Gemini 3.8 Flash 158.8 (#25 of 159), DeepSeek V4 Pro 0813 156.0 (#43 of 159) and Mistral Large 4 154.6 (#50 of 159). The confidence ranges of the top two overlap (156.8–161.0 vs 154.1–158.1), so treat the gap as small. On individual benchmarks: GPQA Diamond — Gemini 3.8 Flash 95.4%, DeepSeek V4 Pro 0813 91.7%, Mistral Large 4 90.2%; OTIS Mock AIME 2024–2025 — Mistral Large 4 99.4%, Gemini 3.8 Flash 98.9%, DeepSeek V4 Pro 0813 98.6%; SimpleQA Verified — Gemini 3.8 Flash 69.7%, DeepSeek V4 Pro 0813 52.9%, Mistral Large 4 22.4%.
Which is better for coding?
There are no published SWE-bench Verified results for DeepSeek V4 Pro 0813, Gemini 3.8 Flash and Mistral Large 4 yet, so there is no like-for-like coding score. On overall capability, Gemini 3.8 Flash leads, which tends to carry over to coding, but test on your own codebase. All three support tool calling for agent workflows.
Which has the bigger context window?
Gemini 3.8 Flash and Mistral Large 4 have the largest context windows (1,048,576 and 1,048,576 tokens), against 1,000,000 for DeepSeek V4 Pro 0813. Maximum output per response: DeepSeek V4 Pro 0813 up to 384,000, Gemini 3.8 Flash up to 65,536, Mistral Large 4 up to 262,144 tokens.
Which can read images, PDFs, audio or video?
DeepSeek V4 Pro 0813 accepts text; Gemini 3.8 Flash accepts text, images, PDFs, audio and video; Mistral Large 4 accepts text and images. Gemini 3.8 Flash handles the widest range of inputs.
Are any of these open source?
DeepSeek V4 Pro 0813 publishes its weights (MIT) and can be self-hosted; Gemini 3.8 Flash and Mistral Large 4 is proprietary.
Which is newer?
Mistral Large 4 is the newest, released Oct 6, 2026. Gemini 3.8 Flash came out Sep 2, 2026; DeepSeek V4 Pro 0813 came out Aug 13, 2026.
How do you decide the winner?
This rating uses capability (50%, independent benchmark results); price (25%, blended price per million tokens (3 input : 1 output), log scale); inputs & features (15%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (10%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Nov 22, 2026.This rating uses capability (67%, independent benchmark results); inputs & features (20%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (13%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Nov 22, 2026.