Codestral 25.08 vs ERNIE 4.5 21B A3B vs ERNIE 4.5 300B A47B
ERNIE 4.5 21B A3B comes out ahead, 57 to 47 and 43 on our weighted score, and it is the cheaper option too.
Mistral AICodestral 25.08
47/100- ECI—
- Price$0.30 / $0.90
- Context128K
- Our pick
BaiduERNIE 4.5 21B A3B
57/100- ECI—
- Price$0.07 / $0.28
- Context131K
- Baidu
ERNIE 4.5 300B A47B
43/100- ECI—
- Price$0.28 / $1.10
- Context131K
ERNIE 4.5 21B A3B is our pick
ERNIE 4.5 21B A3B is the better all-round choice, scoring 57/100 against Codestral 25.08 (47) and ERNIE 4.5 300B A47B (43). It leads on price. Codestral 25.08 wins on inputs & features. The score weighs price 50%, inputs & features 30%, context window 20%. None of the shared benchmarks cover every model here yet, so capability is left out of this verdict.
- CapabilityNot enough dataNo independent benchmark covers every model here yet
- Lowest priceERNIE 4.5 21B A3BERNIE 4.5 21B A3B $0.122 · Codestral 25.08 $0.45 · ERNIE 4.5 300B A47B $0.485 per 1M tokens (3:1 blend)
- Longest contextERNIE 4.5 21B A3B and ERNIE 4.5 300B A47BERNIE 4.5 21B A3B 131,072 · ERNIE 4.5 300B A47B 131,072 · Codestral 25.08 128,000 tokens
- Widest inputsSame inputsCodestral 25.08: Text · ERNIE 4.5 21B A3B: Text · ERNIE 4.5 300B A47B: Text
- Self-hostingERNIE 4.5 21B A3B and ERNIE 4.5 300B A47BPublishes downloadable weights (Apache-2.0)
| Measure | Weight | Codestral 25.08 | ERNIE 4.5 21B A3B | ERNIE 4.5 300B A47B |
|---|---|---|---|---|
| Price | 50% | 64 | 89 | 62 |
| Inputs & features | 30% | 35 | 25 | 25 |
| Context window | 20% | 24 | 25 | 25 |
| Overall | 100% | 47/100 | 57/100 | 43/100 |
Left out because at least one model lacks the data: capability. The remaining weights were rescaled.
Every spec in one table
Highlighted cells lead their row. Dashes mean the data is not published.
| Specification | Codestral 25.08Mistral AI | ERNIE 4.5 21B A3BBaidu | ERNIE 4.5 300B A47BBaidu |
|---|---|---|---|
| Capability | |||
| Capabilities Index (ECI) | — | — | — |
| ECI rank | — | — | — |
| Price per million tokens | |||
| Input | $0.30 | $0.07 (best) | $0.28 |
| Output | $0.90 | $0.28 (best) | $1.10 |
| Cached input | $0.03 | — | — |
| Blended (3:1) | $0.45 | $0.122 (best) | $0.485 |
| Long-context rate | Same rate | Same rate | Same rate |
| Price source | Official Mistral API | Median of 1 providers | Median of 1 providers |
| Limits | |||
| Context window | 128,000 tokens | 131,072 tokens (best) | 131,072 tokens (best) |
| Max output | 8,192 tokens | 8,000 tokens | 12,000 tokens (best) |
| Inputs and features | |||
| Text | Yes | Yes | Yes |
| Images | No | No | No |
| PDFs | No | No | No |
| Audio | No | No | No |
| Video | No | No | No |
| Reasoning | No | No | No |
| Tool calling | Yes | Yes | Yes |
| Structured output | Yes | No | No |
| Availability | |||
| Weights | Proprietary | OpenApache-2.0 | OpenApache-2.0 |
| API model ID | codestral-2508 | — | — |
| API providers | 11 (best) | 1 | 1 |
| Released | Jul 30, 2025 | Jun 30, 2025 | Jun 30, 2025 |
| Knowledge cutoff | — | — | — |
What would a month cost?
Enter your expected volume in millions of tokens. List prices only; caching and batch discounts would lower these.
- Codestral 25.08$4.80
- ERNIE 4.5 21B A3B$1.26
- ERNIE 4.5 300B A47B$5.00
Which should you choose?
Which is better: Codestral 25.08, ERNIE 4.5 21B A3B or ERNIE 4.5 300B A47B?
ERNIE 4.5 21B A3B is the better all-round choice, scoring 57/100 against Codestral 25.08 (47) and ERNIE 4.5 300B A47B (43). It leads on price. Codestral 25.08 wins on inputs & features. The score weighs price 50%, inputs & features 30%, context window 20%. None of the shared benchmarks cover every model here yet, so capability is left out of this verdict.
Which is cheaper, Codestral 25.08, ERNIE 4.5 21B A3B or ERNIE 4.5 300B A47B?
ERNIE 4.5 21B A3B is cheaper at $0.07 input / $0.28 output per million tokens (median across 1 API provider). Codestral 25.08 costs $0.30 input / $0.90 output per million tokens (official Mistral API price); ERNIE 4.5 300B A47B costs $0.28 input / $1.10 output per million tokens (median across 1 API provider). At a typical mix of three input tokens to one output token, that is $0.122 per million tokens for ERNIE 4.5 21B A3B versus $0.45 for Codestral 25.08 (3.7× as much) and $0.485 for ERNIE 4.5 300B A47B (4× as much).
Which scores higher on benchmarks?
There is no independent benchmark that covers all three models yet. Codestral 25.08 has not been scored yet, ERNIE 4.5 21B A3B has not been scored yet and ERNIE 4.5 300B A47B has not been scored yet.
Which is better for coding?
There are no published SWE-bench Verified results for Codestral 25.08, ERNIE 4.5 21B A3B and ERNIE 4.5 300B A47B yet, so there is no like-for-like coding score. Test both on a sample of your own tasks. All three support tool calling for agent workflows.
Which has the bigger context window?
ERNIE 4.5 21B A3B and ERNIE 4.5 300B A47B have the largest context windows (131,072 and 131,072 tokens), against 128,000 for Codestral 25.08. Maximum output per response: Codestral 25.08 up to 8,192, ERNIE 4.5 21B A3B up to 8,000, ERNIE 4.5 300B A47B up to 12,000 tokens.
Which can read images, PDFs, audio or video?
Codestral 25.08 accepts text; ERNIE 4.5 21B A3B accepts text; ERNIE 4.5 300B A47B accepts text. They handle the same number of input types.
Are any of these open source?
ERNIE 4.5 21B A3B and ERNIE 4.5 300B A47B publishes its weights (Apache-2.0) and can be self-hosted; Codestral 25.08 is proprietary.
Which is newer?
Codestral 25.08 is the newest, released Jul 30, 2025. ERNIE 4.5 21B A3B came out Jun 30, 2025; ERNIE 4.5 300B A47B came out Jun 30, 2025.
How do you decide the winner?
Each model gets a 0–100 score on 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 2, 2026.
