Codestral 2508 vs Qwen3 14B
Qwen3 14B costs 3.9x less per output token than Codestral 2508 ($0.23 vs $0.90 per million). Updated October 21, 2026; 1 benchmarks have a result for both.
Codestral 2508 or Qwen3 14B: which to pick
Neither model has a coding score yet, so the two cannot be ranked on coding.
Pick Codestral 2508 for a larger context window (256K against 41K tokens) and work measured by Kagi LLM Benchmark, where it leads.
- Shared results
- 1
- Codestral 2508 only
- 7
- Qwen3 14B only
- 19
- Like-for-like categories
- 1 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, six public coding benchmarks
No call
Neither has a coding score yet.
One side not measured
Intelligence
AA Intelligence Index
No call
Only Qwen3 14B has a AA Intelligence Index result, 6.7.
Directional only
Web design
Design Arena: websites
No call
Only Codestral 2508 has a Design Arena: websites result, 1013.
Directional only
Long documents
Listed context window
Codestral 2508
Codestral 2508 takes 256K tokens against 41K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Qwen3 14B
Codestral 2508 $0.0008, Qwen3 14B $0.0002 per turn.
List prices
At a glance
- Coding index
- Codestral 2508 not benchmarked, Qwen3 14B not benchmarked.
- Widest category
- Reasoning, 16.6 points apart on average over 1 shared benchmark.
- Widest single gap
- Kagi LLM Benchmark: Codestral 2508 32.5% against Qwen3 14B 49.1%, 16.6 points apart on a 0 to 100 scale.
- Output price
- Codestral 2508 $0.90 against Qwen3 14B $0.23 per million output tokens, 3.9x apart.
- Context window
- Codestral 2508 takes 256K tokens, Qwen3 14B 41K.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | Codestral 2508 | Bars | Qwen3 14B | Ties |
|---|---|---|---|---|
| Reasoning | 0 | 1 | ||
| All shared boards | 0 | 1 |
CodingDirectional onlySix of these boards feed the Vibe Coding Index4 boards, none shared
| Benchmark | Codestral 2508 | Qwen3 14B | Leader |
|---|---|---|---|
| LiveCodeBench (Artificial Analysis run) | Not measured | 52.3% | Only Qwen3 14B measured |
| SciCode | Not measured | 30.7% | Only Qwen3 14B measured |
| Terminal-Bench Hard (Artificial Analysis run) | Not measured | 5.3% | Only Qwen3 14B measured |
| ALE-Bench | 137.8 | Not measured | Only Codestral 2508 measured |
| Neither model has a result on the other 35 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | Codestral 2508 | Qwen3 14B | Leader |
|---|---|---|---|
| AA Intelligence Index | Not measured | 6.7 | Only Qwen3 14B measured |
| Epoch Capabilities Index | Not measured | 138.2 | Only Qwen3 14B measured |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | Codestral 2508 | Qwen3 14B | Leader |
|---|---|---|---|
| τ²-Bench Telecom | Not measured | 34.5% | Only Qwen3 14B measured |
| Berkeley Function Calling Leaderboard (BFCL) | Not measured | 41.0% | Only Qwen3 14B measured |
| Neither model has a result on the other 39 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Codestral 2508 | Qwen3 14B | Leader |
|---|---|---|---|
| GPQA Diamond (Artificial Analysis run) | Not measured | 60.4% | Only Qwen3 14B measured |
| Humanity's Last Exam (Artificial Analysis run) | Not measured | 4.5% | Only Qwen3 14B measured |
| GPQA Diamond (Epoch AI run) | Not measured | 63.8% | Only Qwen3 14B measured |
| Kagi LLM Benchmark | 32.5% | 49.1% | Qwen3 14B |
| Neither model has a result on the other 18 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index3 boards, none shared
| Benchmark | Codestral 2508 | Qwen3 14B | Leader |
|---|---|---|---|
| AA Math Index | Not measured | 58.0 | Only Qwen3 14B measured |
| AIME 2025 (Artificial Analysis run) | Not measured | 58.0% | Only Qwen3 14B measured |
| OTIS Mock AIME | Not measured | 66.4% | Only Qwen3 14B measured |
| Neither model has a result on the other 10 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index1 board, none shared
| Benchmark | Codestral 2508 | Qwen3 14B | Leader |
|---|---|---|---|
| MMLU Pro (Artificial Analysis run) | Not measured | 77.4% | Only Qwen3 14B measured |
| Neither model has a result on the other 17 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index6 boards, none shared
| Benchmark | Codestral 2508 | Qwen3 14B | Leader |
|---|---|---|---|
| Design Arena: websites | 1013 | Not measured | Only Codestral 2508 measured |
| Design Arena: UI components | 1017 | Not measured | Only Codestral 2508 measured |
| Design Arena: game dev | 968 | Not measured | Only Codestral 2508 measured |
| Design Arena: data viz | 1017 | Not measured | Only Codestral 2508 measured |
| Design Arena: 3D | 1020 | Not measured | Only Codestral 2508 measured |
| Design Arena: all code categories | 1007 | Not measured | Only Codestral 2508 measured |
| Neither model has a result on the other 3 boards here | |||
GeneralDirectional onlyPublic boards, shown as published, kept out of the index5 boards, none shared
| Benchmark | Codestral 2508 | Qwen3 14B | Leader |
|---|---|---|---|
| IFBench | Not measured | 40.5% | Only Qwen3 14B measured |
| Long context reasoning (AA-LCR) | Not measured | 0.0% | Only Qwen3 14B measured |
| DTBench | Not measured | 64.0% | Only Qwen3 14B measured |
| LMCA | Not measured | 18.2% | Only Qwen3 14B measured |
| Fiction.LiveBench | Not measured | 62.5% | Only Qwen3 14B measured |
| Neither model has a result on the other 41 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Codestral 2508 | Qwen3 14B | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
Qwen3 14B costs less
Repository review
50K input + 3K output tokens
Qwen3 14B costs less
Agent loop
200K input + 10K output tokens
Qwen3 14B costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | Mistral Codestral 2508 | Qwen3 14B |
|---|---|---|
| Maker | Mistral AI | Qwen |
| Weights | No public weights listed | Open weights |
| Released | August 1, 2025 | April 28, 2025 |
| Context window | 256K | 41K |
| Input $/Mtok | $0.30 | $0.11 |
| Output $/Mtok | $0.90 | $0.23 |
| Coding index | Not benchmarked | Not benchmarked |
| Independent measurements behind it | 0 | 0 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Codestral 2508 better than Qwen3 14B for coding?
Neither model has a coding score yet, so the two cannot be ranked on coding. Pick Codestral 2508 for a larger context window (256K against 41K tokens) and work measured by Kagi LLM Benchmark, where it leads.
Which is cheaper, Codestral 2508 or Qwen3 14B?
Codestral 2508 costs $0.30 per million input tokens and $0.90 per million output tokens; Qwen3 14B costs $0.11 and $0.23. Cost per solved task is known for neither of them.
Which has the bigger context window, Codestral 2508 or Qwen3 14B?
Codestral 2508, with 256K tokens against 41K.
How many benchmarks is the Codestral 2508 vs Qwen3 14B comparison based on?
1 benchmarks have a result for both models as of October 21, 2026. Codestral 2508's coding index rests on 0 independent benchmarks and Qwen3 14B's on 0. Where only one model has a result, the page says so instead of naming a winner.